Update install instructions

Add support for using phonemes as direct input

CHANGELOG.md

@@ -0,0 +1,19 @@
+# Changelog
+
+## 3.2.0
+
+- Refactor as `piper_sample_generator` package
+
+## 3.1.0
+
+- Support MPS acceleration on Apple Silicon
+- Add `--phoneme-input` flag
+
+## 3.0.0
+
+- Move phonemization to piper 1.3.0 (piper-phonemize is deprecated)
+- Move to PyTorch 2
+- Add support for using Piper voices (`.onnx`) directly
+- Allow multiple `--model` for Piper voices (`.onnx`)
+- Remove silence trimming
+- Remove `min-phoneme-count`

README.md

@@ -1,35 +1,51 @@
 # Piper Sample Generator
 
-Generates samples using [Piper](https://github.com/rhasspy/piper/) for training a wake word system like [openWakeWord](https://github.com/dscripka/openWakeWord).
+Generate spoken audio samples using [Piper][piper] for training a wake word system like [openWakeWord][] or [microWakeWord][].
 
+Supports normal [Piper voices][piper voices] or a special [generator][] that can mix speaker embeddings (English only).
 
 ## Install
 
-Create a virtual environment and install the requirements:
-
 ``` sh
-git clone https://github.com/rhasspy/piper-sample-generator.git
+pip install piper-sample-generator
-cd piper-sample-generator/
-
-python3 -m venv .venv
-source .venv/bin/activate
-python3 -m pip install --upgrade pip
-python3 -m pip install -r requirements.txt
 ```
 
-Download the LibriTTS-R generator (exported from [checkpoint](https://huggingface.co/datasets/rhasspy/piper-checkpoints/tree/main/en/en_US/libritts_r/medium)):
+## Piper Voices
+
+Download one or more [Piper voices][piper voices] (both the `.onnx` and `.onnx.json` files for each voice). [Audio samples][piper samples] are available.
+
+As an example, we'll download the U.S. English "lessac" voice in medium quality:
+
+``` sh
+mkdir -p voices
+wget -O voices/en_US-lessac-medium.onnx 'https://huggingface.co/rhasspy/piper-voices/resolve/main/en/en_US/lessac/medium/en_US-lessac-medium.onnx?download=true'
+wget -O voices/en_US-lessac-medium.onnx.json 'https://huggingface.co/rhasspy/piper-voices/resolve/main/en/en_US/lessac/medium/en_US-lessac-medium.onnx.json?download=true'
+```
+
+Generate a small set of samples with the CLI:
+
+``` sh
+python3 -m piper_sample_generator 'okay piper.' --model voices/en_US-lessac-medium.onnx --max-samples 10 --output-dir okay_piper/
+```
+
+Check the `okay_piper/` directory for 10 WAV files (named `0.wav` to `9.wav`).
+
+You can add multiple `--model <voice>` arguments to cycle between different voices when generating samples.
+
+See `--help` for more options, including `--length-scales` (speaking speeds).
+
+## Generator
+
+Download the LibriTTS-R generator (exported from [checkpoint][]):
 
 ``` sh
 wget -O models/en-us-libritts-high.pt 'https://github.com/rhasspy/piper-sample-generator/releases/download/v2.0.0/en_US-libritts_r-medium.pt'
 ```
 
-
-## Run
-
 Generate a small set of samples with the CLI:
 
 ``` sh
-python3 generate_samples.py 'okay, piper.' --max-samples 10 --output-dir okay_piper/
+python3 -m piper_sample_generator 'okay piper.' --model models/en-us-libritts-high.pt --max-samples 10 --output-dir okay_piper/
 ```
 
 Check the `okay_piper/` directory for 10 WAV files (named `0.wav` to `9.wav`).
@@ -37,38 +53,38 @@ Check the `okay_piper/` directory for 10 WAV files (named `0.wav` to `9.wav`).
 Generation can be much faster and more efficient if you have a GPU available and PyTorch is configured to use it. In this case, increase the batch size:
 
 ``` sh
-python3 generate_samples.py 'okay, piper.' --max-samples 100 --batch-size 10 --output-dir okay_piper/
+python3 -m piper_sample_generator 'okay piper.' --model models/en-us-libritts-high.pt --max-samples 100 --batch-size 10 --output-dir okay_piper/
 ```
 
 On an NVidia 2080 Ti with 11GB, a batch size of 100 was possible (generating approximately 100 samples per second).
 
 Setting `--max-speakers` to a value less than 904 (the number of speakers LibriTTS) is recommended. Because very few samples of later speakers were in the original dataset, using them can cause audio artifacts.
 
-See `--help` for more options, including adjust the `--length-scales` (speaking speeds) and `--slerp-weights` (speaker blending) which are cycled per batch.
+See `--help` for more options, including the `--length-scales` (speaking speeds) and `--slerp-weights` (speaker blending) which are cycled per batch.
-
-Alternatively, you can import the generate function into another Python script:
-
-```python
-from generate_samples import generate_samples  # make sure to add this to your Python path as needed
-
-generate_samples(text = ["okay, piper"], max_samples = 100, output_dir = output_dir, batch_size=10)
-```
-
-There are some additional arguments available when importing the function directly, see the docstring of `generate_sample` for more information.
 
 ### Augmentation
 
-Once you have samples generating, you can augment them using [audiomentation](https://iver56.github.io/audiomentations/):
+Once you have samples generated, you can augment them using [audiomentation](https://iver56.github.io/audiomentations/):
 
 ``` sh
-python3 augment.py --sample-rate 16000 okay_piper/ okay_piper_augmented/
+python3 -m piper_sample_generator.augment --sample-rate 22050 okay_piper/ okay_piper_augmented/
 ```
 
 This will do several things to each sample:
 
 1. Randomly decrease the volume
     * The original samples are normalized, so different volume levels are needed
-2. Randomly [apply an impulse response](https://iver56.github.io/audiomentations/waveform_transforms/apply_impulse_response/) using the files in `impulses/`
+2. Randomly apply an [impulse response][] using the files in `piper_sample_generator/impulses/`
     * Change the acoustics of the sample to sound like the speaker was in a room with echo or using a poor quality microphone
-3. Resample to 16Khz for training (e.g., [openWakeWord](https://github.com/dscripka/openWakeWord))
+3. Resample to 16Khz for training (e.g., [openWakeWord][])
 
+
+<!-- Links -->
+[piper]: https://github.com/OHF-Voice/piper1-gpl/
+[openWakeWord]: https://github.com/dscripka/openWakeWord
+[microWakeWord]: https://github.com/kahrendt/microWakeWord/
+[piper voices]: https://huggingface.co/rhasspy/piper-voices
+[generator]: https://github.com/rhasspy/piper-sample-generator/releases/download/v2.0.0/en_US-libritts_r-medium.pt
+[piper samples]: https://rhasspy.github.io/piper-samples/
+[checkpoint]: https://huggingface.co/datasets/rhasspy/piper-checkpoints/tree/main/en/en_US/libritts_r/medium
+[impulse response]: https://iver56.github.io/audiomentations/waveform_transforms/apply_impulse_response/

generate_samples.py

@@ -1,407 +0,0 @@
-#!/usr/bin/env python3
-import argparse
-import gc
-import itertools as it
-import json
-import logging
-import os
-import wave
-from pathlib import Path
-from typing import List, Union
-
-import numpy as np
-import torch
-import torchaudio
-import webrtcvad
-from piper_phonemize import phonemize_espeak
-
-from piper_train.vits import commons
-
-_DIR = Path(__file__).parent
-_LOGGER = logging.getLogger(__name__)
-logging.basicConfig(level=logging.DEBUG)
-
-
-# Main generation function
-def generate_samples(
-    text: Union[List, str],
-    output_dir: str,
-    max_samples: int = None,
-    file_names: List[str] = [],
-    model: str = os.path.join(
-        Path(__file__).parent, "models", "en_US-libritts_r-medium.pt"
-    ),
-    batch_size: int = 1,
-    slerp_weights: List[float] = [0.5],
-    length_scales: List[float] = [0.75, 1, 1.25],
-    noise_scales: List[float] = [0.667],
-    noise_scale_ws: List[float] = [0.8],
-    max_speakers: float = None,
-    verbose: bool = False,
-    auto_reduce_batch_size: bool = False,
-    **kwargs,
-) -> None:
-    """
-    Generate synthetic speech clips, saving the clips to the specified output directory.
-
-    Args:
-        text (List[str]): The text to convert into speech. Can be either a
-                          a list of strings, or a path to a file with text on each line.
-        output_dir (str): The location to save the generated clips.
-        max_samples (int): The maximum number of samples to generate.
-        file_names (List[str]): The names to use when saving the files. Must be the same length
-                                as the `text` argument, if a list.
-        model (str): The path to the STT model to use for generation.
-        batch_size (int): The batch size to use when generated the clips
-        slerp_weights (List[float]): The weights to use when mixing speakers via SLERP.
-        length_scales (List[float]): Controls the average duration/speed of the generated speech.
-        noise_scales (List[float]): A parameter for overall variability of the generated speech.
-        noise_scale_ws (List[float]): A parameter for the stochastic duration of words/phonemes.
-        max_speakers (int): The maximum speaker number to use, if the model is multi-speaker.
-        verbose (bool): Enable or disable more detailed logging messages (default: False).
-        auto_reduce_batch_size (bool): Automatically and temporarily reduce the batch size
-                                       if CUDA OOM errors are detected, and try to resume generation.
-
-    Returns:
-        None
-    """
-
-    if max_samples is None:
-        max_samples = len(text)
-
-    _LOGGER.debug("Loading %s", model)
-    model_path = Path(model)
-    model = torch.load(model_path)
-    model.eval()
-    _LOGGER.info("Successfully loaded the model")
-
-    if torch.cuda.is_available():
-        model.cuda()
-        _LOGGER.debug("CUDA available, using GPU")
-
-    output_dir = Path(output_dir)
-    output_dir.mkdir(parents=True, exist_ok=True)
-
-    config_path = f"{model_path}.json"
-    with open(config_path, "r", encoding="utf-8") as config_file:
-        config = json.load(config_file)
-
-    voice = config["espeak"]["voice"]
-    sample_rate = config["audio"]["sample_rate"]
-    num_speakers = config["num_speakers"]
-    if max_speakers is not None:
-        num_speakers = min(num_speakers, max_speakers)
-
-    max_len = None
-
-    sample_idx = 0
-    is_done = False
-    settings_iter = it.cycle(
-        it.product(
-            slerp_weights,
-            length_scales,
-            noise_scales,
-            noise_scale_ws,
-        )
-    )
-
-    # Define resampler to get to 16khz (https://pytorch.org/audio/stable/tutorials/audio_resampling_tutorial.html#kaiser-best)
-    sample_rate = 22050
-    resample_rate = 16000
-    resampler = torchaudio.transforms.Resample(
-        sample_rate,
-        resample_rate,
-        lowpass_filter_width=64,
-        rolloff=0.9475937167399596,
-        resampling_method="kaiser_window",
-        beta=14.769656459379492,
-    )
-
-    speakers_iter = it.cycle(it.product(range(num_speakers), range(num_speakers)))
-    speakers_batch = list(it.islice(speakers_iter, 0, batch_size))
-    if isinstance(text, str) and os.path.exists(text):
-        texts = it.cycle(
-            [
-                i.strip()
-                for i in open(text, "r", encoding="utf-8").readlines()
-                if len(i.strip()) > 0
-            ]
-        )
-    elif isinstance(text, list):
-        texts = it.cycle(text)
-    else:
-        texts = it.cycle([text])
-
-    if file_names:
-        file_names = it.cycle(file_names)
-
-    batch_idx = 0
-    while speakers_batch:
-        if is_done:
-            break
-
-        batch_size = len(speakers_batch)
-        slerp_weight, length_scale, noise_scale, noise_scale_w = next(settings_iter)
-
-        with torch.no_grad():
-            speaker_1 = torch.LongTensor([s[0] for s in speakers_batch])
-            speaker_2 = torch.LongTensor([s[1] for s in speakers_batch])
-
-            phoneme_ids = [
-                get_phonemes(voice, config, next(texts), verbose)
-                for i in range(batch_size)
-            ]
-
-            def right_pad_lists(lists):
-                max_length = max(len(lst) for lst in lists)
-                padded_lists = []
-                for lst in lists:
-                    padded_l = lst + [1] * (
-                        max_length - len(lst)
-                    )  # phoneme 1 (corresponding to '^' character seems to work best)
-                    padded_lists.append(padded_l)
-                return padded_lists
-
-            phoneme_ids = right_pad_lists(phoneme_ids)
-
-            if auto_reduce_batch_size:
-                oom_error = True
-                counter = 1
-                while oom_error is True:
-                    try:
-                        audio = generate_audio(
-                            model,
-                            speaker_1[0 : batch_size // counter],
-                            speaker_2[0 : batch_size // counter],
-                            phoneme_ids[0 : batch_size // counter],
-                            slerp_weight,
-                            noise_scale,
-                            noise_scale_w,
-                            length_scale,
-                            max_len,
-                        )
-                        oom_error = False
-                    except torch.cuda.OutOfMemoryError:
-                        torch.cuda.empty_cache()
-                        gc.collect()
-                        counter += 1  # reduce batch size to avoid OOM errors
-            else:
-                audio = generate_audio(
-                    model,
-                    speaker_1,
-                    speaker_2,
-                    phoneme_ids,
-                    slerp_weight,
-                    noise_scale,
-                    noise_scale_w,
-                    length_scale,
-                    max_len,
-                )
-
-            # Resample audio
-            audio = resampler(audio.cpu()).numpy()
-
-            audio_int16 = audio_float_to_int16(audio)
-            for audio_idx in range(audio_int16.shape[0]):
-                # Use webrtcvad to trip silence from the clips
-                audio_data = remove_silence(audio_int16[audio_idx].flatten())[None,]
-
-                if isinstance(file_names, it.cycle):
-                    wav_path = output_dir / next(file_names)
-                else:
-                    wav_path = output_dir / f"{sample_idx}.wav"
-                with wave.open(str(wav_path), "wb") as wav_file:
-                    wav_file.setframerate(resample_rate)
-                    wav_file.setsampwidth(2)
-                    wav_file.setnchannels(1)
-                    wav_file.writeframes(audio_data)
-
-                sample_idx += 1
-                if sample_idx >= max_samples:
-                    is_done = True
-                    break
-
-            # print(f"Batch {batch_idx +1}/{max_samples//batch_size} complete", " "*200, end='\r')
-
-        # Next batch
-        _LOGGER.debug(f"Batch {batch_idx +1}/{max_samples//batch_size} complete")
-        speakers_batch = list(it.islice(speakers_iter, 0, batch_size))
-        batch_idx += 1
-
-    _LOGGER.info("Done")
-
-
-def remove_silence(x, frame_duration=0.030, sample_rate=16000, min_start=2000):
-    """Uses webrtc voice activity detection to remove silence from the clips"""
-    vad = webrtcvad.Vad(0)
-    if x.dtype == np.float32 or x.dtype == np.float64:
-        x = (x * 32767).astype(np.int16)
-    x_new = x[0:min_start].tolist()
-    step_size = int(sample_rate * frame_duration)
-    for i in range(min_start, x.shape[0] - step_size, step_size):
-        vad_res = vad.is_speech(x[i : i + step_size].tobytes(), sample_rate)
-        if vad_res:
-            x_new.extend(x[i : i + step_size].tolist())
-    return np.array(x_new).astype(np.int16)
-
-
-def generate_audio(
-    model,
-    speaker_1,
-    speaker_2,
-    phoneme_ids,
-    slerp_weight,
-    noise_scale,
-    noise_scale_w,
-    length_scale,
-    max_len,
-):
-    x = torch.LongTensor(phoneme_ids)
-    x_lengths = torch.LongTensor([len(i) for i in phoneme_ids])
-
-    if torch.cuda.is_available():
-        speaker_1 = speaker_1.cuda()
-        speaker_2 = speaker_2.cuda()
-        x = x.cuda()
-        x_lengths = x_lengths.cuda()
-
-    x, m_p_orig, logs_p_orig, x_mask = model.enc_p(x, x_lengths)
-    emb0 = model.emb_g(speaker_1)
-    emb1 = model.emb_g(speaker_2)
-    g = slerp(emb0, emb1, slerp_weight).unsqueeze(-1)  # [b, h, 1]
-
-    if model.use_sdp:
-        logw = model.dp(x, x_mask, g=g, reverse=True, noise_scale=noise_scale_w)
-    else:
-        logw = model.dp(x, x_mask, g=g)
-    w = torch.exp(logw) * x_mask * length_scale
-    w_ceil = torch.ceil(w)
-    y_lengths = torch.clamp_min(torch.sum(w_ceil, [1, 2]), 1).long()
-    y_mask = torch.unsqueeze(
-        commons.sequence_mask(y_lengths, y_lengths.max()), 1
-    ).type_as(x_mask)
-    attn_mask = torch.unsqueeze(x_mask, 2) * torch.unsqueeze(y_mask, -1)
-    attn = commons.generate_path(w_ceil, attn_mask)
-
-    m_p = torch.matmul(attn.squeeze(1), m_p_orig.transpose(1, 2)).transpose(
-        1, 2
-    )  # [b, t', t], [b, t, d] -> [b, d, t']
-    logs_p = torch.matmul(attn.squeeze(1), logs_p_orig.transpose(1, 2)).transpose(
-        1, 2
-    )  # [b, t', t], [b, t, d] -> [b, d, t']
-
-    z_p = m_p + torch.randn_like(m_p) * torch.exp(logs_p) * noise_scale
-    z = model.flow(z_p, y_mask, g=g, reverse=True)
-    o = model.dec((z * y_mask)[:, :, :max_len], g=g)
-
-    audio = o
-    return audio
-
-
-def get_phonemes(voice, config, text, verbose):
-    # Combine all sentences
-    phonemes = [
-        p
-        for sentence_phonemes in phonemize_espeak(text, voice)
-        for p in sentence_phonemes
-    ]
-    if verbose is True:
-        _LOGGER.debug("Phonemes: %s", phonemes)
-
-    id_map = config["phoneme_id_map"]
-    phoneme_ids = list(id_map["^"])
-    for phoneme in phonemes:
-        p_ids = id_map.get(phoneme)
-        if p_ids is not None:
-            phoneme_ids.extend(p_ids)
-            phoneme_ids.extend(id_map["_"])
-
-    phoneme_ids.extend(id_map["$"])
-    return phoneme_ids
-
-
-def slerp(v1, v2, t, DOT_THR=0.9995, zdim=-1):
-    """SLERP for pytorch tensors interpolating `v1` to `v2` with scale of `t`.
-
-    `DOT_THR` determines when the vectors are too close to parallel.
-        If they are too close, then a regular linear interpolation is used.
-
-    `zdim` is the feature dimension over which to compute norms and find angles.
-        For example: if a sequence of 5 vectors is input with shape [5, 768]
-        Then `zdim = 1` or `zdim = -1` computes SLERP along the feature dim of 768.
-
-    Theory Reference:
-    https://splines.readthedocs.io/en/latest/rotation/slerp.html
-    PyTorch reference:
-    https://discuss.pytorch.org/t/help-regarding-slerp-function-for-generative-model-sampling/32475/3
-    Numpy reference:
-    https://gist.github.com/dvschultz/3af50c40df002da3b751efab1daddf2c
-    """
-
-    # take the dot product between normalized vectors
-    v1_norm = v1 / torch.norm(v1, dim=zdim, keepdim=True)
-    v2_norm = v2 / torch.norm(v2, dim=zdim, keepdim=True)
-    dot = (v1_norm * v2_norm).sum(zdim)
-
-    # if the vectors are too close, return a simple linear interpolation
-    if (torch.abs(dot) > DOT_THR).any():
-        res = (1 - t) * v1 + t * v2
-
-    # else apply SLERP
-    else:
-        # compute the angle terms we need
-        theta = torch.acos(dot)
-        theta_t = theta * t
-        sin_theta = torch.sin(theta)
-        sin_theta_t = torch.sin(theta_t)
-
-        # compute the sine scaling terms for the vectors
-        s1 = torch.sin(theta - theta_t) / sin_theta
-        s2 = sin_theta_t / sin_theta
-
-        # interpolate the vectors
-        res = (s1.unsqueeze(zdim) * v1) + (s2.unsqueeze(zdim) * v2)
-
-    return res
-
-
-def audio_float_to_int16(
-    audio: np.ndarray, max_wav_value: float = 32767.0
-) -> np.ndarray:
-    """Normalize audio and convert to int16 range"""
-    audio_norm = audio * (max_wav_value / max(0.01, np.max(np.abs(audio))))
-    audio_norm = np.clip(audio_norm, -max_wav_value, max_wav_value)
-    audio_norm = audio_norm.astype("int16")
-    return audio_norm
-
-
-if __name__ == "__main__":
-    # Get command line arguments
-    parser = argparse.ArgumentParser()
-    parser.add_argument("text")
-    parser.add_argument("--max-samples", required=True, type=int)
-    parser.add_argument(
-        "--model", default=_DIR / "models" / "en_US-libritts_r-medium.pt"
-    )
-    parser.add_argument("--batch-size", type=int, default=1)
-    parser.add_argument("--slerp-weights", nargs="+", type=float, default=[0.5])
-    parser.add_argument(
-        "--length-scales", nargs="+", type=float, default=[1.0, 0.75, 1.25, 1.4]
-    )
-    parser.add_argument(
-        "--noise-scales",
-        nargs="+",
-        type=float,
-        default=[0.667, 0.75, 0.85, 0.9, 1.0, 1.4],
-    )
-    parser.add_argument("--noise-scale-ws", nargs="+", type=float, default=[0.8])
-    parser.add_argument("--output-dir", default="output")
-    parser.add_argument(
-        "--max-speakers",
-        type=int,
-        help="Maximum number of speakers to use (default: all)",
-    )
-    args = parser.parse_args().__dict__
-
-    # Generate speech
-    generate_samples(**args)

models/de_DE-mls-medium.pt.json

@@ -0,0 +1,740 @@
+{
+    "dataset": "mls",
+    "audio": {
+        "sample_rate": 22050,
+        "quality": "medium"
+    },
+    "espeak": {
+        "voice": "de"
+    },
+    "language": {
+        "code": "de_DE"
+    },
+    "inference": {
+        "noise_scale": 0.333,
+        "length_scale": 1,
+        "noise_w": 0.333
+    },
+    "phoneme_type": "espeak",
+    "phoneme_map": {},
+    "phoneme_id_map": {
+        " ": [
+            3
+        ],
+        "!": [
+            4
+        ],
+        "\"": [
+            150
+        ],
+        "#": [
+            149
+        ],
+        "$": [
+            2
+        ],
+        "'": [
+            5
+        ],
+        "(": [
+            6
+        ],
+        ")": [
+            7
+        ],
+        ",": [
+            8
+        ],
+        "-": [
+            9
+        ],
+        ".": [
+            10
+        ],
+        "0": [
+            130
+        ],
+        "1": [
+            131
+        ],
+        "2": [
+            132
+        ],
+        "3": [
+            133
+        ],
+        "4": [
+            134
+        ],
+        "5": [
+            135
+        ],
+        "6": [
+            136
+        ],
+        "7": [
+            137
+        ],
+        "8": [
+            138
+        ],
+        "9": [
+            139
+        ],
+        ":": [
+            11
+        ],
+        ";": [
+            12
+        ],
+        "?": [
+            13
+        ],
+        "X": [
+            156
+        ],
+        "^": [
+            1
+        ],
+        "_": [
+            0
+        ],
+        "a": [
+            14
+        ],
+        "b": [
+            15
+        ],
+        "c": [
+            16
+        ],
+        "d": [
+            17
+        ],
+        "e": [
+            18
+        ],
+        "f": [
+            19
+        ],
+        "g": [
+            154
+        ],
+        "h": [
+            20
+        ],
+        "i": [
+            21
+        ],
+        "j": [
+            22
+        ],
+        "k": [
+            23
+        ],
+        "l": [
+            24
+        ],
+        "m": [
+            25
+        ],
+        "n": [
+            26
+        ],
+        "o": [
+            27
+        ],
+        "p": [
+            28
+        ],
+        "q": [
+            29
+        ],
+        "r": [
+            30
+        ],
+        "s": [
+            31
+        ],
+        "t": [
+            32
+        ],
+        "u": [
+            33
+        ],
+        "v": [
+            34
+        ],
+        "w": [
+            35
+        ],
+        "x": [
+            36
+        ],
+        "y": [
+            37
+        ],
+        "z": [
+            38
+        ],
+        "æ": [
+            39
+        ],
+        "ç": [
+            40
+        ],
+        "ð": [
+            41
+        ],
+        "ø": [
+            42
+        ],
+        "ħ": [
+            43
+        ],
+        "ŋ": [
+            44
+        ],
+        "œ": [
+            45
+        ],
+        "ǀ": [
+            46
+        ],
+        "ǁ": [
+            47
+        ],
+        "ǂ": [
+            48
+        ],
+        "ǃ": [
+            49
+        ],
+        "ɐ": [
+            50
+        ],
+        "ɑ": [
+            51
+        ],
+        "ɒ": [
+            52
+        ],
+        "ɓ": [
+            53
+        ],
+        "ɔ": [
+            54
+        ],
+        "ɕ": [
+            55
+        ],
+        "ɖ": [
+            56
+        ],
+        "ɗ": [
+            57
+        ],
+        "ɘ": [
+            58
+        ],
+        "ə": [
+            59
+        ],
+        "ɚ": [
+            60
+        ],
+        "ɛ": [
+            61
+        ],
+        "ɜ": [
+            62
+        ],
+        "ɞ": [
+            63
+        ],
+        "ɟ": [
+            64
+        ],
+        "ɠ": [
+            65
+        ],
+        "ɡ": [
+            66
+        ],
+        "ɢ": [
+            67
+        ],
+        "ɣ": [
+            68
+        ],
+        "ɤ": [
+            69
+        ],
+        "ɥ": [
+            70
+        ],
+        "ɦ": [
+            71
+        ],
+        "ɧ": [
+            72
+        ],
+        "ɨ": [
+            73
+        ],
+        "ɪ": [
+            74
+        ],
+        "ɫ": [
+            75
+        ],
+        "ɬ": [
+            76
+        ],
+        "ɭ": [
+            77
+        ],
+        "ɮ": [
+            78
+        ],
+        "ɯ": [
+            79
+        ],
+        "ɰ": [
+            80
+        ],
+        "ɱ": [
+            81
+        ],
+        "ɲ": [
+            82
+        ],
+        "ɳ": [
+            83
+        ],
+        "ɴ": [
+            84
+        ],
+        "ɵ": [
+            85
+        ],
+        "ɶ": [
+            86
+        ],
+        "ɸ": [
+            87
+        ],
+        "ɹ": [
+            88
+        ],
+        "ɺ": [
+            89
+        ],
+        "ɻ": [
+            90
+        ],
+        "ɽ": [
+            91
+        ],
+        "ɾ": [
+            92
+        ],
+        "ʀ": [
+            93
+        ],
+        "ʁ": [
+            94
+        ],
+        "ʂ": [
+            95
+        ],
+        "ʃ": [
+            96
+        ],
+        "ʄ": [
+            97
+        ],
+        "ʈ": [
+            98
+        ],
+        "ʉ": [
+            99
+        ],
+        "ʊ": [
+            100
+        ],
+        "ʋ": [
+            101
+        ],
+        "ʌ": [
+            102
+        ],
+        "ʍ": [
+            103
+        ],
+        "ʎ": [
+            104
+        ],
+        "ʏ": [
+            105
+        ],
+        "ʐ": [
+            106
+        ],
+        "ʑ": [
+            107
+        ],
+        "ʒ": [
+            108
+        ],
+        "ʔ": [
+            109
+        ],
+        "ʕ": [
+            110
+        ],
+        "ʘ": [
+            111
+        ],
+        "ʙ": [
+            112
+        ],
+        "ʛ": [
+            113
+        ],
+        "ʜ": [
+            114
+        ],
+        "ʝ": [
+            115
+        ],
+        "ʟ": [
+            116
+        ],
+        "ʡ": [
+            117
+        ],
+        "ʢ": [
+            118
+        ],
+        "ʦ": [
+            155
+        ],
+        "ʰ": [
+            145
+        ],
+        "ʲ": [
+            119
+        ],
+        "ˈ": [
+            120
+        ],
+        "ˌ": [
+            121
+        ],
+        "ː": [
+            122
+        ],
+        "ˑ": [
+            123
+        ],
+        "˞": [
+            124
+        ],
+        "ˤ": [
+            146
+        ],
+        "̃": [
+            141
+        ],
+        "̊": [
+            158
+        ],
+        "̝": [
+            157
+        ],
+        "̧": [
+            140
+        ],
+        "̩": [
+            144
+        ],
+        "̪": [
+            142
+        ],
+        "̯": [
+            143
+        ],
+        "̺": [
+            152
+        ],
+        "̻": [
+            153
+        ],
+        "β": [
+            125
+        ],
+        "ε": [
+            147
+        ],
+        "θ": [
+            126
+        ],
+        "χ": [
+            127
+        ],
+        "ᵻ": [
+            128
+        ],
+        "↑": [
+            151
+        ],
+        "↓": [
+            148
+        ],
+        "ⱱ": [
+            129
+        ]
+    },
+    "num_symbols": 256,
+    "num_speakers": 236,
+    "speaker_id_map": {
+        "2422": 0,
+        "4536": 1,
+        "2037": 2,
+        "9565": 3,
+        "10148": 4,
+        "6507": 5,
+        "5055": 6,
+        "3503": 7,
+        "252": 8,
+        "9132": 9,
+        "3990": 10,
+        "5753": 11,
+        "5424": 12,
+        "2602": 13,
+        "4174": 14,
+        "3885": 15,
+        "12415": 16,
+        "8470": 17,
+        "11927": 18,
+        "9639": 19,
+        "3494": 20,
+        "2946": 21,
+        "5283": 22,
+        "4533": 23,
+        "2497": 24,
+        "12275": 25,
+        "1649": 26,
+        "146": 27,
+        "8337": 28,
+        "4542": 29,
+        "589": 30,
+        "1998": 31,
+        "3797": 32,
+        "5244": 33,
+        "7328": 34,
+        "7998": 35,
+        "10179": 36,
+        "9610": 37,
+        "20": 38,
+        "253": 39,
+        "12899": 40,
+        "7194": 41,
+        "3759": 42,
+        "2677": 43,
+        "6719": 44,
+        "1897": 45,
+        "11990": 46,
+        "6880": 47,
+        "19": 48,
+        "9515": 49,
+        "327": 50,
+        "3244": 51,
+        "5324": 52,
+        "2234": 53,
+        "3124": 54,
+        "2043": 55,
+        "143": 56,
+        "8139": 57,
+        "9646": 58,
+        "8659": 59,
+        "9538": 60,
+        "989": 61,
+        "5405": 62,
+        "10087": 63,
+        "8294": 64,
+        "4396": 65,
+        "1474": 66,
+        "139": 67,
+        "136": 68,
+        "10791": 69,
+        "7242": 70,
+        "3631": 71,
+        "9908": 72,
+        "7906": 73,
+        "1171": 74,
+        "7479": 75,
+        "5632": 76,
+        "3731": 77,
+        "4650": 78,
+        "135": 79,
+        "145": 80,
+        "137": 81,
+        "1757": 82,
+        "91": 83,
+        "9514": 84,
+        "13494": 85,
+        "1946": 86,
+        "3277": 87,
+        "5595": 88,
+        "278": 89,
+        "7120": 90,
+        "7406": 91,
+        "11695": 92,
+        "1593": 93,
+        "3862": 94,
+        "138": 95,
+        "141": 96,
+        "9948": 97,
+        "1163": 98,
+        "1054": 99,
+        "1844": 100,
+        "4911": 101,
+        "7261": 102,
+        "8223": 103,
+        "7624": 104,
+        "144": 105,
+        "13871": 106,
+        "2974": 107,
+        "5934": 108,
+        "7002": 109,
+        "8769": 110,
+        "3363": 111,
+        "3040": 112,
+        "6067": 113,
+        "9494": 114,
+        "8743": 115,
+        "13255": 116,
+        "1660": 117,
+        "3588": 118,
+        "4748": 119,
+        "8450": 120,
+        "5295": 121,
+        "4705": 122,
+        "8125": 123,
+        "7272": 124,
+        "7320": 125,
+        "1874": 126,
+        "1262": 127,
+        "10870": 128,
+        "12379": 129,
+        "4463": 130,
+        "10349": 131,
+        "2252": 132,
+        "8325": 133,
+        "3052": 134,
+        "4001": 135,
+        "7456": 136,
+        "140": 137,
+        "4739": 138,
+        "11299": 139,
+        "4730": 140,
+        "6659": 141,
+        "2034": 142,
+        "2732": 143,
+        "2158": 144,
+        "3698": 145,
+        "5675": 146,
+        "6315": 147,
+        "10904": 148,
+        "7202": 149,
+        "11480": 150,
+        "10625": 151,
+        "11546": 152,
+        "4576": 153,
+        "4512": 154,
+        "8634": 155,
+        "13626": 156,
+        "1613": 157,
+        "287": 158,
+        "9207": 159,
+        "6982": 160,
+        "1724": 161,
+        "10191": 162,
+        "1091": 163,
+        "2909": 164,
+        "1965": 165,
+        "2506": 166,
+        "4414": 167,
+        "5764": 168,
+        "12776": 169,
+        "1033": 170,
+        "13726": 171,
+        "2314": 172,
+        "6826": 173,
+        "9706": 174,
+        "8427": 175,
+        "9168": 176,
+        "9287": 177,
+        "6905": 178,
+        "4153": 179,
+        "3330": 180,
+        "2859": 181,
+        "5406": 182,
+        "2840": 183,
+        "1920": 184,
+        "9241": 185,
+        "10163": 186,
+        "8305": 187,
+        "12461": 188,
+        "3276": 189,
+        "11413": 190,
+        "10536": 191,
+        "10614": 192,
+        "7579": 193,
+        "8675": 194,
+        "7483": 195,
+        "7270": 196,
+        "8704": 197,
+        "4468": 198,
+        "6611": 199,
+        "11497": 200,
+        "11772": 201,
+        "2792": 202,
+        "11481": 203,
+        "10162": 204,
+        "10819": 205,
+        "8732": 206,
+        "11328": 207,
+        "11920": 208,
+        "6646": 209,
+        "7486": 210,
+        "11870": 211,
+        "12417": 212,
+        "10364": 213,
+        "6117": 214,
+        "6448": 215,
+        "10433": 216,
+        "7515": 217,
+        "5823": 218,
+        "8567": 219,
+        "10947": 220,
+        "11869": 221,
+        "12335": 222,
+        "12500": 223,
+        "13755": 224,
+        "7006": 225,
+        "3685": 226,
+        "5819": 227,
+        "9353": 228,
+        "11355": 229,
+        "12174": 230,
+        "7150": 231,
+        "6952": 232,
+        "11987": 233,
+        "3995": 234,
+        "7449": 235
+    },
+    "piper_version": "1.0.0"
+}

models/fr_FR-mls-medium.pt.json

@@ -0,0 +1,629 @@
+{
+    "dataset": "mls",
+    "audio": {
+        "sample_rate": 22050,
+        "quality": "medium"
+    },
+    "espeak": {
+        "voice": "fr"
+    },
+    "language": {
+        "code": "fr_FR"
+    },
+    "inference": {
+        "noise_scale": 0.333,
+        "length_scale": 1,
+        "noise_w": 0.333
+    },
+    "phoneme_type": "espeak",
+    "phoneme_map": {},
+    "phoneme_id_map": {
+        " ": [
+            3
+        ],
+        "!": [
+            4
+        ],
+        "\"": [
+            150
+        ],
+        "#": [
+            149
+        ],
+        "$": [
+            2
+        ],
+        "'": [
+            5
+        ],
+        "(": [
+            6
+        ],
+        ")": [
+            7
+        ],
+        ",": [
+            8
+        ],
+        "-": [
+            9
+        ],
+        ".": [
+            10
+        ],
+        "0": [
+            130
+        ],
+        "1": [
+            131
+        ],
+        "2": [
+            132
+        ],
+        "3": [
+            133
+        ],
+        "4": [
+            134
+        ],
+        "5": [
+            135
+        ],
+        "6": [
+            136
+        ],
+        "7": [
+            137
+        ],
+        "8": [
+            138
+        ],
+        "9": [
+            139
+        ],
+        ":": [
+            11
+        ],
+        ";": [
+            12
+        ],
+        "?": [
+            13
+        ],
+        "X": [
+            156
+        ],
+        "^": [
+            1
+        ],
+        "_": [
+            0
+        ],
+        "a": [
+            14
+        ],
+        "b": [
+            15
+        ],
+        "c": [
+            16
+        ],
+        "d": [
+            17
+        ],
+        "e": [
+            18
+        ],
+        "f": [
+            19
+        ],
+        "g": [
+            154
+        ],
+        "h": [
+            20
+        ],
+        "i": [
+            21
+        ],
+        "j": [
+            22
+        ],
+        "k": [
+            23
+        ],
+        "l": [
+            24
+        ],
+        "m": [
+            25
+        ],
+        "n": [
+            26
+        ],
+        "o": [
+            27
+        ],
+        "p": [
+            28
+        ],
+        "q": [
+            29
+        ],
+        "r": [
+            30
+        ],
+        "s": [
+            31
+        ],
+        "t": [
+            32
+        ],
+        "u": [
+            33
+        ],
+        "v": [
+            34
+        ],
+        "w": [
+            35
+        ],
+        "x": [
+            36
+        ],
+        "y": [
+            37
+        ],
+        "z": [
+            38
+        ],
+        "æ": [
+            39
+        ],
+        "ç": [
+            40
+        ],
+        "ð": [
+            41
+        ],
+        "ø": [
+            42
+        ],
+        "ħ": [
+            43
+        ],
+        "ŋ": [
+            44
+        ],
+        "œ": [
+            45
+        ],
+        "ǀ": [
+            46
+        ],
+        "ǁ": [
+            47
+        ],
+        "ǂ": [
+            48
+        ],
+        "ǃ": [
+            49
+        ],
+        "ɐ": [
+            50
+        ],
+        "ɑ": [
+            51
+        ],
+        "ɒ": [
+            52
+        ],
+        "ɓ": [
+            53
+        ],
+        "ɔ": [
+            54
+        ],
+        "ɕ": [
+            55
+        ],
+        "ɖ": [
+            56
+        ],
+        "ɗ": [
+            57
+        ],
+        "ɘ": [
+            58
+        ],
+        "ə": [
+            59
+        ],
+        "ɚ": [
+            60
+        ],
+        "ɛ": [
+            61
+        ],
+        "ɜ": [
+            62
+        ],
+        "ɞ": [
+            63
+        ],
+        "ɟ": [
+            64
+        ],
+        "ɠ": [
+            65
+        ],
+        "ɡ": [
+            66
+        ],
+        "ɢ": [
+            67
+        ],
+        "ɣ": [
+            68
+        ],
+        "ɤ": [
+            69
+        ],
+        "ɥ": [
+            70
+        ],
+        "ɦ": [
+            71
+        ],
+        "ɧ": [
+            72
+        ],
+        "ɨ": [
+            73
+        ],
+        "ɪ": [
+            74
+        ],
+        "ɫ": [
+            75
+        ],
+        "ɬ": [
+            76
+        ],
+        "ɭ": [
+            77
+        ],
+        "ɮ": [
+            78
+        ],
+        "ɯ": [
+            79
+        ],
+        "ɰ": [
+            80
+        ],
+        "ɱ": [
+            81
+        ],
+        "ɲ": [
+            82
+        ],
+        "ɳ": [
+            83
+        ],
+        "ɴ": [
+            84
+        ],
+        "ɵ": [
+            85
+        ],
+        "ɶ": [
+            86
+        ],
+        "ɸ": [
+            87
+        ],
+        "ɹ": [
+            88
+        ],
+        "ɺ": [
+            89
+        ],
+        "ɻ": [
+            90
+        ],
+        "ɽ": [
+            91
+        ],
+        "ɾ": [
+            92
+        ],
+        "ʀ": [
+            93
+        ],
+        "ʁ": [
+            94
+        ],
+        "ʂ": [
+            95
+        ],
+        "ʃ": [
+            96
+        ],
+        "ʄ": [
+            97
+        ],
+        "ʈ": [
+            98
+        ],
+        "ʉ": [
+            99
+        ],
+        "ʊ": [
+            100
+        ],
+        "ʋ": [
+            101
+        ],
+        "ʌ": [
+            102
+        ],
+        "ʍ": [
+            103
+        ],
+        "ʎ": [
+            104
+        ],
+        "ʏ": [
+            105
+        ],
+        "ʐ": [
+            106
+        ],
+        "ʑ": [
+            107
+        ],
+        "ʒ": [
+            108
+        ],
+        "ʔ": [
+            109
+        ],
+        "ʕ": [
+            110
+        ],
+        "ʘ": [
+            111
+        ],
+        "ʙ": [
+            112
+        ],
+        "ʛ": [
+            113
+        ],
+        "ʜ": [
+            114
+        ],
+        "ʝ": [
+            115
+        ],
+        "ʟ": [
+            116
+        ],
+        "ʡ": [
+            117
+        ],
+        "ʢ": [
+            118
+        ],
+        "ʦ": [
+            155
+        ],
+        "ʰ": [
+            145
+        ],
+        "ʲ": [
+            119
+        ],
+        "ˈ": [
+            120
+        ],
+        "ˌ": [
+            121
+        ],
+        "ː": [
+            122
+        ],
+        "ˑ": [
+            123
+        ],
+        "˞": [
+            124
+        ],
+        "ˤ": [
+            146
+        ],
+        "̃": [
+            141
+        ],
+        "̊": [
+            158
+        ],
+        "̝": [
+            157
+        ],
+        "̧": [
+            140
+        ],
+        "̩": [
+            144
+        ],
+        "̪": [
+            142
+        ],
+        "̯": [
+            143
+        ],
+        "̺": [
+            152
+        ],
+        "̻": [
+            153
+        ],
+        "β": [
+            125
+        ],
+        "ε": [
+            147
+        ],
+        "θ": [
+            126
+        ],
+        "χ": [
+            127
+        ],
+        "ᵻ": [
+            128
+        ],
+        "↑": [
+            151
+        ],
+        "↓": [
+            148
+        ],
+        "ⱱ": [
+            129
+        ]
+    },
+    "num_symbols": 256,
+    "num_speakers": 125,
+    "speaker_id_map": {
+        "1840": 0,
+        "3698": 1,
+        "123": 2,
+        "1474": 3,
+        "12709": 4,
+        "7423": 5,
+        "9242": 6,
+        "8778": 7,
+        "3060": 8,
+        "4512": 9,
+        "6249": 10,
+        "12541": 11,
+        "13634": 12,
+        "10065": 13,
+        "6128": 14,
+        "5232": 15,
+        "5764": 16,
+        "12713": 17,
+        "12823": 18,
+        "6070": 19,
+        "12501": 20,
+        "9121": 21,
+        "1649": 22,
+        "2776": 23,
+        "11772": 24,
+        "5612": 25,
+        "11822": 26,
+        "1590": 27,
+        "5525": 28,
+        "10827": 29,
+        "1243": 30,
+        "13142": 31,
+        "62": 32,
+        "13177": 33,
+        "10620": 34,
+        "8102": 35,
+        "8582": 36,
+        "11875": 37,
+        "7239": 38,
+        "9854": 39,
+        "7377": 40,
+        "10082": 41,
+        "12512": 42,
+        "1329": 43,
+        "2506": 44,
+        "6856": 45,
+        "10058": 46,
+        "103": 47,
+        "14": 48,
+        "6381": 49,
+        "1664": 50,
+        "11954": 51,
+        "66": 52,
+        "1127": 53,
+        "3270": 54,
+        "13611": 55,
+        "13658": 56,
+        "12968": 57,
+        "1989": 58,
+        "12981": 59,
+        "7193": 60,
+        "6348": 61,
+        "7679": 62,
+        "2284": 63,
+        "3182": 64,
+        "3503": 65,
+        "2033": 66,
+        "2771": 67,
+        "7614": 68,
+        "125": 69,
+        "3204": 70,
+        "5595": 71,
+        "5553": 72,
+        "694": 73,
+        "1624": 74,
+        "1887": 75,
+        "2926": 76,
+        "7150": 77,
+        "3190": 78,
+        "3344": 79,
+        "4699": 80,
+        "1798": 81,
+        "1745": 82,
+        "5077": 83,
+        "753": 84,
+        "52": 85,
+        "4174": 86,
+        "4018": 87,
+        "12899": 88,
+        "1844": 89,
+        "4396": 90,
+        "1817": 91,
+        "2155": 92,
+        "2946": 93,
+        "4336": 94,
+        "4609": 95,
+        "1977": 96,
+        "10957": 97,
+        "204": 98,
+        "4650": 99,
+        "5295": 100,
+        "5968": 101,
+        "4744": 102,
+        "2825": 103,
+        "9804": 104,
+        "707": 105,
+        "30": 106,
+        "115": 107,
+        "5840": 108,
+        "2587": 109,
+        "2607": 110,
+        "2544": 111,
+        "28": 112,
+        "27": 113,
+        "177": 114,
+        "112": 115,
+        "94": 116,
+        "2596": 117,
+        "3595": 118,
+        "7032": 119,
+        "7848": 120,
+        "11247": 121,
+        "7439": 122,
+        "2904": 123,
+        "6362": 124
+    },
+    "piper_version": "1.0.0"
+}

models/nl_NL-mls-medium.pt.json

@@ -0,0 +1,554 @@
+{
+    "audio": {
+        "sample_rate": 22050
+    },
+    "espeak": {
+        "voice": "nl"
+    },
+    "language": {
+        "code": "nl_NL"
+    },
+    "inference": {
+        "noise_scale": 0.333,
+        "length_scale": 1,
+        "noise_w": 0.333
+    },
+    "phoneme_type": "espeak",
+    "phoneme_map": {},
+    "phoneme_id_map": {
+        " ": [
+            3
+        ],
+        "!": [
+            4
+        ],
+        "\"": [
+            150
+        ],
+        "#": [
+            149
+        ],
+        "$": [
+            2
+        ],
+        "'": [
+            5
+        ],
+        "(": [
+            6
+        ],
+        ")": [
+            7
+        ],
+        ",": [
+            8
+        ],
+        "-": [
+            9
+        ],
+        ".": [
+            10
+        ],
+        "0": [
+            130
+        ],
+        "1": [
+            131
+        ],
+        "2": [
+            132
+        ],
+        "3": [
+            133
+        ],
+        "4": [
+            134
+        ],
+        "5": [
+            135
+        ],
+        "6": [
+            136
+        ],
+        "7": [
+            137
+        ],
+        "8": [
+            138
+        ],
+        "9": [
+            139
+        ],
+        ":": [
+            11
+        ],
+        ";": [
+            12
+        ],
+        "?": [
+            13
+        ],
+        "X": [
+            156
+        ],
+        "^": [
+            1
+        ],
+        "_": [
+            0
+        ],
+        "a": [
+            14
+        ],
+        "b": [
+            15
+        ],
+        "c": [
+            16
+        ],
+        "d": [
+            17
+        ],
+        "e": [
+            18
+        ],
+        "f": [
+            19
+        ],
+        "g": [
+            154
+        ],
+        "h": [
+            20
+        ],
+        "i": [
+            21
+        ],
+        "j": [
+            22
+        ],
+        "k": [
+            23
+        ],
+        "l": [
+            24
+        ],
+        "m": [
+            25
+        ],
+        "n": [
+            26
+        ],
+        "o": [
+            27
+        ],
+        "p": [
+            28
+        ],
+        "q": [
+            29
+        ],
+        "r": [
+            30
+        ],
+        "s": [
+            31
+        ],
+        "t": [
+            32
+        ],
+        "u": [
+            33
+        ],
+        "v": [
+            34
+        ],
+        "w": [
+            35
+        ],
+        "x": [
+            36
+        ],
+        "y": [
+            37
+        ],
+        "z": [
+            38
+        ],
+        "æ": [
+            39
+        ],
+        "ç": [
+            40
+        ],
+        "ð": [
+            41
+        ],
+        "ø": [
+            42
+        ],
+        "ħ": [
+            43
+        ],
+        "ŋ": [
+            44
+        ],
+        "œ": [
+            45
+        ],
+        "ǀ": [
+            46
+        ],
+        "ǁ": [
+            47
+        ],
+        "ǂ": [
+            48
+        ],
+        "ǃ": [
+            49
+        ],
+        "ɐ": [
+            50
+        ],
+        "ɑ": [
+            51
+        ],
+        "ɒ": [
+            52
+        ],
+        "ɓ": [
+            53
+        ],
+        "ɔ": [
+            54
+        ],
+        "ɕ": [
+            55
+        ],
+        "ɖ": [
+            56
+        ],
+        "ɗ": [
+            57
+        ],
+        "ɘ": [
+            58
+        ],
+        "ə": [
+            59
+        ],
+        "ɚ": [
+            60
+        ],
+        "ɛ": [
+            61
+        ],
+        "ɜ": [
+            62
+        ],
+        "ɞ": [
+            63
+        ],
+        "ɟ": [
+            64
+        ],
+        "ɠ": [
+            65
+        ],
+        "ɡ": [
+            66
+        ],
+        "ɢ": [
+            67
+        ],
+        "ɣ": [
+            68
+        ],
+        "ɤ": [
+            69
+        ],
+        "ɥ": [
+            70
+        ],
+        "ɦ": [
+            71
+        ],
+        "ɧ": [
+            72
+        ],
+        "ɨ": [
+            73
+        ],
+        "ɪ": [
+            74
+        ],
+        "ɫ": [
+            75
+        ],
+        "ɬ": [
+            76
+        ],
+        "ɭ": [
+            77
+        ],
+        "ɮ": [
+            78
+        ],
+        "ɯ": [
+            79
+        ],
+        "ɰ": [
+            80
+        ],
+        "ɱ": [
+            81
+        ],
+        "ɲ": [
+            82
+        ],
+        "ɳ": [
+            83
+        ],
+        "ɴ": [
+            84
+        ],
+        "ɵ": [
+            85
+        ],
+        "ɶ": [
+            86
+        ],
+        "ɸ": [
+            87
+        ],
+        "ɹ": [
+            88
+        ],
+        "ɺ": [
+            89
+        ],
+        "ɻ": [
+            90
+        ],
+        "ɽ": [
+            91
+        ],
+        "ɾ": [
+            92
+        ],
+        "ʀ": [
+            93
+        ],
+        "ʁ": [
+            94
+        ],
+        "ʂ": [
+            95
+        ],
+        "ʃ": [
+            96
+        ],
+        "ʄ": [
+            97
+        ],
+        "ʈ": [
+            98
+        ],
+        "ʉ": [
+            99
+        ],
+        "ʊ": [
+            100
+        ],
+        "ʋ": [
+            101
+        ],
+        "ʌ": [
+            102
+        ],
+        "ʍ": [
+            103
+        ],
+        "ʎ": [
+            104
+        ],
+        "ʏ": [
+            105
+        ],
+        "ʐ": [
+            106
+        ],
+        "ʑ": [
+            107
+        ],
+        "ʒ": [
+            108
+        ],
+        "ʔ": [
+            109
+        ],
+        "ʕ": [
+            110
+        ],
+        "ʘ": [
+            111
+        ],
+        "ʙ": [
+            112
+        ],
+        "ʛ": [
+            113
+        ],
+        "ʜ": [
+            114
+        ],
+        "ʝ": [
+            115
+        ],
+        "ʟ": [
+            116
+        ],
+        "ʡ": [
+            117
+        ],
+        "ʢ": [
+            118
+        ],
+        "ʦ": [
+            155
+        ],
+        "ʰ": [
+            145
+        ],
+        "ʲ": [
+            119
+        ],
+        "ˈ": [
+            120
+        ],
+        "ˌ": [
+            121
+        ],
+        "ː": [
+            122
+        ],
+        "ˑ": [
+            123
+        ],
+        "˞": [
+            124
+        ],
+        "ˤ": [
+            146
+        ],
+        "̃": [
+            141
+        ],
+        "̊": [
+            158
+        ],
+        "̝": [
+            157
+        ],
+        "̧": [
+            140
+        ],
+        "̩": [
+            144
+        ],
+        "̪": [
+            142
+        ],
+        "̯": [
+            143
+        ],
+        "̺": [
+            152
+        ],
+        "̻": [
+            153
+        ],
+        "β": [
+            125
+        ],
+        "ε": [
+            147
+        ],
+        "θ": [
+            126
+        ],
+        "χ": [
+            127
+        ],
+        "ᵻ": [
+            128
+        ],
+        "↑": [
+            151
+        ],
+        "↓": [
+            148
+        ],
+        "ⱱ": [
+            129
+        ]
+    },
+    "num_symbols": 256,
+    "num_speakers": 52,
+    "speaker_id_map": {
+        "2450": 0,
+        "1724": 1,
+        "1666": 2,
+        "5809": 3,
+        "496": 4,
+        "2506": 5,
+        "7432": 6,
+        "3619": 7,
+        "4429": 8,
+        "3798": 9,
+        "12500": 10,
+        "10587": 11,
+        "2951": 12,
+        "1775": 13,
+        "9861": 14,
+        "880": 15,
+        "3034": 16,
+        "2825": 17,
+        "5438": 18,
+        "3245": 19,
+        "4396": 20,
+        "11290": 21,
+        "11936": 22,
+        "6916": 23,
+        "10294": 24,
+        "10079": 25,
+        "7588": 26,
+        "7579": 27,
+        "123": 28,
+        "3024": 29,
+        "960": 30,
+        "10984": 31,
+        "2792": 32,
+        "7723": 33,
+        "4174": 34,
+        "2981": 35,
+        "5764": 36,
+        "6513": 37,
+        "7884": 38,
+        "6697": 39,
+        "12749": 40,
+        "11157": 41,
+        "2239": 42,
+        "10879": 43,
+        "1085": 44,
+        "8480": 45,
+        "8331": 46,
+        "6282": 47,
+        "10632": 48,
+        "2602": 49,
+        "5367": 50,
+        "11472": 51
+    },
+    "piper_version": "1.0.0"
+}

piper_sample_generator/__init__.py

@@ -0,0 +1 @@
+"""Piper sample generator."""

piper_sample_generator/__main__.py

@@ -0,0 +1,628 @@
+#!/usr/bin/env python3
+import argparse
+import gc
+import itertools as it
+import json
+import logging
+import os
+import unicodedata
+import wave
+from collections.abc import Iterable
+from pathlib import Path
+from typing import Any, Dict, List, Optional, Tuple, Union, cast
+
+import numpy as np
+import torch
+from piper import PiperVoice, SynthesisConfig
+from piper.phonemize_espeak import EspeakPhonemizer
+
+try:
+    from piper_train.vits import commons
+except ImportError:
+    from piper_train.vits import commons
+
+_LOGGER = logging.getLogger(__name__)
+logging.basicConfig(level=logging.DEBUG)
+
+
+# Main generation function
+def generate_samples(
+    text: Union[List[str], str],
+    output_dir: Union[str, Path],
+    model: Union[str, Path],
+    max_samples: Optional[int] = None,
+    file_names: Optional[Iterable[str]] = None,
+    batch_size: int = 1,
+    slerp_weights: Tuple[float, ...] = (0.5,),
+    length_scales: Tuple[float, ...] = (0.75, 1, 1.25),
+    noise_scales: Tuple[float, ...] = (0.667,),
+    noise_scale_ws: Tuple[float, ...] = (0.8,),
+    max_speakers: Optional[int] = None,
+    verbose: bool = False,
+    phoneme_input: bool = False,
+    **kwargs,
+) -> None:
+    """
+    Generate synthetic speech clips, saving the clips to the specified output directory.
+
+    Args:
+        text (List[str]): The text to convert into speech. Can be either a
+                          a list of strings, or a path to a file with text on each line.
+        output_dir (str): The location to save the generated clips.
+        model (str): The path to the TTS generator model (.pt).
+        max_samples (int): The maximum number of samples to generate.
+        file_names (List[str]): The names to use when saving the files. Must be the same length
+                                as the `text` argument, if a list.
+        batch_size (int): The batch size to use when generated the clips
+        slerp_weights (List[float]): The weights to use when mixing speakers via SLERP.
+        length_scales (List[float]): Controls the average duration/speed of the generated speech.
+        noise_scales (List[float]): A parameter for overall variability of the generated speech.
+        noise_scale_ws (List[float]): A parameter for the stochastic duration of words/phonemes.
+        max_speakers (int): The maximum speaker number to use, if the model is multi-speaker.
+        verbose (bool): Enable or disable more detailed logging messages (default: False).
+        phoneme_input (bool): Set to indicate given input text is phoneme input.
+    Returns:
+        None
+    """
+
+    if max_samples is None:
+        max_samples = len(text)
+
+    _LOGGER.debug("Loading %s", model)
+    model_path = Path(model)
+
+    torch_model = torch.load(model_path, weights_only=False)
+    torch_model.eval()
+    _LOGGER.info("Successfully loaded the model")
+
+    if torch.cuda.is_available():
+        torch_model.cuda()
+        _LOGGER.debug("CUDA available, using GPU")
+    elif torch.backends.mps.is_available():
+        mps_device = torch.device("mps")
+        torch_model.to(mps_device)
+        _LOGGER.debug("MPS available, using GPU")
+
+    output_dir = Path(output_dir)
+    output_dir.mkdir(parents=True, exist_ok=True)
+
+    config_path = f"{model_path}.json"
+    with open(config_path, "r", encoding="utf-8") as config_file:
+        config = json.load(config_file)
+
+    voice = config["espeak"]["voice"]
+    sample_rate = config["audio"]["sample_rate"]
+    num_speakers = config["num_speakers"]
+    if max_speakers is not None:
+        num_speakers = min(num_speakers, max_speakers)
+
+    max_len = None
+
+    sample_idx = 0
+    is_done = False
+    settings_iter = it.cycle(
+        it.product(
+            slerp_weights,
+            length_scales,
+            noise_scales,
+            noise_scale_ws,
+        )
+    )
+
+    speakers_iter = it.cycle(it.product(range(num_speakers), range(num_speakers)))
+    speakers_batch = list(it.islice(speakers_iter, 0, batch_size))
+    if isinstance(text, str) and os.path.isfile(text):
+        texts = it.cycle(
+            [
+                i.strip()
+                for i in open(text, "r", encoding="utf-8").readlines()
+                if len(i.strip()) > 0
+            ]
+        )
+    elif isinstance(text, list):
+        texts = it.cycle(text)
+    else:
+        texts = it.cycle([text])
+
+    if file_names:
+        file_names = it.cycle(file_names)
+
+    batch_idx = 0
+    while speakers_batch:
+        if is_done:
+            break
+
+        batch_size = len(speakers_batch)
+        slerp_weight, length_scale, noise_scale, noise_scale_w = next(settings_iter)
+
+        with torch.no_grad():
+            speaker_1 = torch.LongTensor([s[0] for s in speakers_batch])
+            speaker_2 = torch.LongTensor([s[1] for s in speakers_batch])
+
+            phoneme_ids_by_batch = []
+            for i in range(batch_size):
+                phoneme_ids = get_phonemes(
+                    voice, config, next(texts), verbose, phoneme_input
+                )
+                phoneme_ids_by_batch.append(phoneme_ids)
+
+            def right_pad_lists(lists):
+                max_length = max(len(lst) for lst in lists)
+                padded_lists = []
+                for lst in lists:
+                    padded_l = lst + [1] * (
+                        max_length - len(lst)
+                    )  # phoneme 1 (corresponding to '^' character seems to work best)
+                    padded_lists.append(padded_l)
+                return padded_lists
+
+            phoneme_ids_by_batch = right_pad_lists(phoneme_ids_by_batch)
+            audio, phoneme_samples = generate_audio(
+                torch_model,
+                speaker_1,
+                speaker_2,
+                phoneme_ids_by_batch,
+                slerp_weight,
+                noise_scale,
+                noise_scale_w,
+                length_scale,
+                max_len,
+            )
+
+            # Trim audio to actual length based on phoneme samples
+            for i in range(audio.shape[0]):
+                # Fill time after last speech with silence (zeros)
+                # It will be removed in the next stage with np.trim_zeros
+                last_sample_idx = int(phoneme_samples[i].flatten().sum().item())
+                audio[i, 0, last_sample_idx + 1 :] = 0
+
+            audio_numpy = audio.cpu().numpy()
+
+            if torch.backends.mps.is_available():
+                # There seems to be a memory leak if we don't empty the cache
+                # after each batch with mps
+                torch.mps.empty_cache()
+                gc.collect()
+
+            audio_int16 = audio_float_to_int16(audio_numpy)
+            for audio_idx in range(audio_int16.shape[0]):
+                audio_data = np.trim_zeros(audio_int16[audio_idx].flatten())
+
+                if isinstance(file_names, it.cycle):
+                    wav_path = output_dir / next(file_names)
+                else:
+                    wav_path = output_dir / f"{sample_idx}.wav"
+
+                wav_file: wave.Wave_write = wave.open(str(wav_path), "wb")
+                with wav_file:
+                    wav_file.setframerate(sample_rate)
+                    wav_file.setsampwidth(2)
+                    wav_file.setnchannels(1)
+                    wav_file.writeframes(audio_data)
+
+                sample_idx += 1
+                if sample_idx >= max_samples:
+                    is_done = True
+                    break
+
+            # print(f"Batch {batch_idx +1}/{max_samples//batch_size} complete", " "*200, end='\r')
+
+        # Next batch
+        _LOGGER.debug("Batch %s/%s complete", batch_idx + 1, max_samples // batch_size)
+        speakers_batch = list(it.islice(speakers_iter, 0, batch_size))
+        batch_idx += 1
+
+    _LOGGER.info("Done")
+
+
+# -----------------------------------------------------------------------------
+
+
+def generate_samples_onnx(
+    text: Union[List[str], str],
+    output_dir: Union[str, Path],
+    model: Union[str, Path, List[Union[str, Path]]],
+    max_samples: Optional[int] = None,
+    file_names: Optional[Iterable[str]] = None,
+    length_scales: Tuple[float, ...] = (0.75, 1, 1.25),
+    noise_scales: Tuple[float, ...] = (0.667,),
+    noise_scale_ws: Tuple[float, ...] = (0.8,),
+    max_speakers: Optional[int] = None,
+    phoneme_input: bool = False,
+    **kwargs,
+) -> None:
+    """
+    Generate synthetic speech clips, saving the clips to the specified output directory.
+
+    Args:
+        text (List[str]): The text to convert into speech. Can be either a
+                          a list of strings, or a path to a file with text on each line.
+        output_dir (str): The location to save the generated clips.
+        model (str): The path to the Piper TTS model (.onnx).
+        max_samples (int): The maximum number of samples to generate.
+        file_names (List[str]): The names to use when saving the files. Must be the same length
+                                as the `text` argument, if a list.
+        length_scales (List[float]): Controls the average duration/speed of the generated speech.
+        noise_scales (List[float]): A parameter for overall variability of the generated speech.
+        noise_scale_ws (List[float]): A parameter for the stochastic duration of words/phonemes.
+        max_speakers (int): The maximum speaker number to use, if the model is multi-speaker.
+        phoneme_input (bool): Set to indicate given input text is phoneme input.
+
+    Returns:
+        None
+    """
+
+    if max_samples is None:
+        max_samples = len(text)
+
+    if not isinstance(model, list):
+        model = [model]
+
+    _LOGGER.debug("Loading %s", model)
+    voices = [PiperVoice.load(m, use_cuda=torch.cuda.is_available()) for m in model]
+    _LOGGER.info("Successfully loaded model(s)")
+
+    output_dir = Path(output_dir)
+    output_dir.mkdir(parents=True, exist_ok=True)
+
+    sample_idx = 0
+    settings_iter = it.cycle(
+        it.product(
+            voices,
+            length_scales,
+            noise_scales,
+            noise_scale_ws,
+        )
+    )
+
+    if isinstance(text, str) and os.path.exists(text):
+        texts = it.cycle(
+            [
+                i.strip()
+                for i in open(text, "r", encoding="utf-8").readlines()
+                if len(i.strip()) > 0
+            ]
+        )
+    elif isinstance(text, list):
+        texts = it.cycle(text)
+    else:
+        texts = it.cycle([text])
+
+    if file_names:
+        file_names = it.cycle(file_names)
+
+    for voice, length_scale, noise_scale, noise_w_scale in settings_iter:
+        num_speakers = voice.config.num_speakers
+        if max_speakers is not None:
+            num_speakers = min(num_speakers, max_speakers)
+
+        for speaker_id in range(num_speakers):
+            if isinstance(file_names, it.cycle):
+                wav_path = output_dir / next(file_names)
+            else:
+                wav_path = output_dir / f"{sample_idx}.wav"
+
+            text_input = next(texts)
+
+            if phoneme_input:
+                # For ONNX models with phoneme input, build phoneme IDs manually
+                phonemes = list(unicodedata.normalize("NFD", text_input))
+
+                # Build phoneme IDs similar to get_phonemes function
+                id_map = voice.config.phoneme_id_map
+
+                # Beginning of utterance
+                phoneme_ids = list(id_map.get("^", [1]))  # Default to [1] if not found
+                phoneme_ids.extend(id_map.get("_", [0]))  # Default to [0] if not found
+
+                # Add phonemes
+                for phoneme in phonemes:
+                    p_ids = id_map.get(phoneme)
+                    if p_ids is not None:
+                        phoneme_ids.extend(p_ids)
+                        phoneme_ids.extend(id_map.get("_", [0]))
+                    else:
+                        _LOGGER.warning(
+                            "Phoneme '%s' not found in model's phoneme map", phoneme
+                        )
+
+                # End of utterance
+                phoneme_ids.extend(id_map.get("$", [2]))  # Default to [2] if not found
+
+                # Generate audio from phoneme IDs
+                syn_config = SynthesisConfig(
+                    speaker_id=speaker_id,
+                    length_scale=length_scale,
+                    noise_scale=noise_scale,
+                    noise_w_scale=noise_w_scale,
+                )
+                audio = voice.phoneme_ids_to_audio(phoneme_ids, syn_config)
+
+                # Convert to int16 and write to WAV
+                audio_int16 = audio_float_to_int16(audio[np.newaxis, :])
+                wav_file: wave.Wave_write = wave.open(str(wav_path), "wb")
+                with wav_file:
+                    wav_file.setframerate(voice.config.sample_rate)
+                    wav_file.setsampwidth(2)
+                    wav_file.setnchannels(1)
+                    wav_file.writeframes(audio_int16.flatten())
+            else:
+                with wave.open(str(wav_path), "wb") as wav_file:
+                    voice.synthesize_wav(
+                        text_input,
+                        wav_file=wav_file,
+                        syn_config=SynthesisConfig(
+                            speaker_id=speaker_id,
+                            length_scale=length_scale,
+                            noise_scale=noise_scale,
+                            noise_w_scale=noise_w_scale,
+                        ),
+                    )
+
+            sample_idx += 1
+            if sample_idx >= max_samples:
+                return
+
+    _LOGGER.info("Done")
+
+
+# -----------------------------------------------------------------------------
+
+
+def generate_audio(
+    model,
+    speaker_1,
+    speaker_2,
+    phoneme_ids,
+    slerp_weight,
+    noise_scale,
+    noise_scale_w,
+    length_scale,
+    max_len,
+) -> Tuple[torch.FloatTensor, torch.FloatTensor]:
+    x = torch.LongTensor(phoneme_ids)
+    x_lengths = torch.LongTensor([len(i) for i in phoneme_ids])
+
+    if torch.cuda.is_available():
+        speaker_1 = speaker_1.cuda()
+        speaker_2 = speaker_2.cuda()
+        x = cast(torch.LongTensor, x.cuda())
+        x_lengths = cast(torch.LongTensor, x_lengths.cuda())
+    elif torch.backends.mps.is_available():
+        mps_device = torch.device("mps")
+        speaker_1 = speaker_1.to(mps_device)
+        speaker_2 = speaker_2.to(mps_device)
+        x = cast(torch.LongTensor, x.to(mps_device))
+        x_lengths = cast(torch.LongTensor, x_lengths.to(mps_device))
+
+    x, m_p_orig, logs_p_orig, x_mask = model.enc_p(x, x_lengths)
+    emb0 = model.emb_g(speaker_1)
+    emb1 = model.emb_g(speaker_2)
+    g = slerp(emb0, emb1, slerp_weight).unsqueeze(-1)  # [b, h, 1]
+
+    if model.use_sdp:
+        logw = model.dp(x, x_mask, g=g, reverse=True, noise_scale=noise_scale_w)
+    else:
+        logw = model.dp(x, x_mask, g=g)
+    w = torch.exp(logw) * x_mask * length_scale
+    w_ceil = torch.ceil(w)
+    y_lengths = torch.clamp_min(torch.sum(w_ceil, [1, 2]), 1).long()
+    y_mask = torch.unsqueeze(
+        commons.sequence_mask(y_lengths, int(y_lengths.max().item())), 1
+    ).type_as(x_mask)
+    attn_mask = torch.unsqueeze(x_mask, 2) * torch.unsqueeze(y_mask, -1)
+    attn = commons.generate_path(w_ceil, attn_mask)
+
+    m_p = torch.matmul(attn.squeeze(1), m_p_orig.transpose(1, 2)).transpose(
+        1, 2
+    )  # [b, t', t], [b, t, d] -> [b, d, t']
+    logs_p = torch.matmul(attn.squeeze(1), logs_p_orig.transpose(1, 2)).transpose(
+        1, 2
+    )  # [b, t', t], [b, t, d] -> [b, d, t']
+
+    z_p = m_p + torch.randn_like(m_p) * torch.exp(logs_p) * noise_scale
+    z = model.flow(z_p, y_mask, g=g, reverse=True)
+    o = model.dec((z * y_mask)[:, :, :max_len], g=g)
+
+    audio = cast(torch.FloatTensor, o)
+    phoneme_samples = cast(torch.FloatTensor, w_ceil * 256)  # hop length
+
+    return audio, phoneme_samples
+
+
+_PHONEMIZER = EspeakPhonemizer()
+
+
+def get_phonemes(
+    voice: str,
+    config: Dict[str, Any],
+    text: str,
+    verbose: bool = False,
+    phoneme_input: bool = False,
+) -> List[int]:
+    # Combine all sentences
+    if phoneme_input:
+        phonemes = list(unicodedata.normalize("NFD", text))
+    else:
+        phonemes = [
+            p
+            for sentence_phonemes in _PHONEMIZER.phonemize(voice, text)
+            for p in sentence_phonemes
+        ]
+    if verbose is True:
+        _LOGGER.debug("Phonemes: %s", phonemes)
+
+    id_map = config["phoneme_id_map"]
+
+    # Beginning of utterance
+    phoneme_ids = list(id_map["^"])
+    phoneme_ids.extend(id_map["_"])
+
+    # Phoneme ids for just the text
+    text_phoneme_ids = []
+
+    for phoneme in phonemes:
+        p_ids = id_map.get(phoneme)
+        if p_ids is not None:
+            phoneme_ids.extend(p_ids)
+            text_phoneme_ids.extend(p_ids)
+            phoneme_ids.extend(id_map["_"])
+            text_phoneme_ids.extend(id_map["_"])
+
+    # End of utterance
+    phoneme_ids.extend(id_map["$"])
+
+    return phoneme_ids
+
+
+def slerp(v1, v2, t: float, DOT_THR: float = 0.9995, zdim: int = -1):
+    """SLERP for pytorch tensors interpolating `v1` to `v2` with scale of `t`.
+
+    `DOT_THR` determines when the vectors are too close to parallel.
+        If they are too close, then a regular linear interpolation is used.
+
+    `zdim` is the feature dimension over which to compute norms and find angles.
+        For example: if a sequence of 5 vectors is input with shape [5, 768]
+        Then `zdim = 1` or `zdim = -1` computes SLERP along the feature dim of 768.
+
+    Theory Reference:
+    https://splines.readthedocs.io/en/latest/rotation/slerp.html
+    PyTorch reference:
+    https://discuss.pytorch.org/t/help-regarding-slerp-function-for-generative-model-sampling/32475/3
+    Numpy reference:
+    https://gist.github.com/dvschultz/3af50c40df002da3b751efab1daddf2c
+    """
+
+    # take the dot product between normalized vectors
+    v1_norm = v1 / torch.norm(v1, dim=zdim, keepdim=True)
+    v2_norm = v2 / torch.norm(v2, dim=zdim, keepdim=True)
+    dot = (v1_norm * v2_norm).sum(zdim)
+
+    # if the vectors are too close, return a simple linear interpolation
+    if (torch.abs(dot) > DOT_THR).any():
+        res = (1 - t) * v1 + t * v2
+
+    # else apply SLERP
+    else:
+        # compute the angle terms we need
+        theta = torch.acos(dot)
+        theta_t = theta * t
+        sin_theta = torch.sin(theta)
+        sin_theta_t = torch.sin(theta_t)
+
+        # compute the sine scaling terms for the vectors
+        s1 = torch.sin(theta - theta_t) / sin_theta
+        s2 = sin_theta_t / sin_theta
+
+        # interpolate the vectors
+        res = (s1.unsqueeze(zdim) * v1) + (s2.unsqueeze(zdim) * v2)
+
+    return res
+
+
+def audio_float_to_int16(
+    audio: np.ndarray, max_wav_value: float = 32767.0
+) -> np.ndarray:
+    """Normalize audio and convert to int16 range"""
+    audio_norm = audio * (max_wav_value / max(0.01, np.max(np.abs(audio))))
+    audio_norm = np.clip(audio_norm, -max_wav_value, max_wav_value)
+    audio_norm = audio_norm.astype("int16")
+    return audio_norm
+
+
+# -----------------------------------------------------------------------------
+
+
+def main() -> int:
+    """Main entry point."""
+
+    # Get command line arguments
+    parser = argparse.ArgumentParser()
+    parser.add_argument("text")
+    parser.add_argument(
+        "--max-samples",
+        required=True,
+        type=int,
+        help="Maximum number of samples to generate",
+    )
+    parser.add_argument(
+        "--model",
+        required=True,
+        action="append",
+        help="Path to PyTorch generator (.pt) or Piper voice model (.onnx)",
+    )
+    parser.add_argument(
+        "--batch-size", type=int, default=1, help="CUDA batch size (generator only)"
+    )
+    parser.add_argument(
+        "--slerp-weights",
+        nargs="+",
+        type=float,
+        default=[0.5],
+        help="Speaker blending weights (generator only)",
+    )
+    parser.add_argument(
+        "--length-scales",
+        nargs="+",
+        type=float,
+        default=[1.0, 0.75, 1.25, 1.4],
+        help="Audio length scales (< 1 is faster, > 1 is slower)",
+    )
+    parser.add_argument(
+        "--noise-scales",
+        nargs="+",
+        type=float,
+        default=[0.667, 0.75, 0.85, 0.9, 1.0, 1.4],
+        help="Noise amounts added to audio (most voices use 0.667)",
+    )
+    parser.add_argument(
+        "--noise-scale-ws",
+        nargs="+",
+        type=float,
+        default=[0.8],
+        help="Phoneme width variation (most voices use 0.8)",
+    )
+    parser.add_argument(
+        "--output-dir",
+        default="output",
+        help="Directory to output WAV files (default: ./output)",
+    )
+    parser.add_argument(
+        "--max-speakers",
+        type=int,
+        help="Maximum number of speakers to use (default: no limit)",
+    )
+    parser.add_argument(
+        "--phoneme-input", action="store_true", help="Treat input text as phoneme input"
+    )
+    parser.add_argument("--verbose", action="store_true")
+    args = parser.parse_args().__dict__
+
+    # Generate speech
+    model_paths = [Path(m) for m in args["model"]]
+    assert model_paths
+
+    if any(mp for mp in model_paths[1:] if mp.suffix != model_paths[0].suffix):
+        _LOGGER.error("All models must have the same suffix (.pt or .onnx)")
+        return 1
+
+    if model_paths[0].suffix == ".onnx":
+        # Use Piper voice (.onnx)
+        generate_samples_onnx(**args)
+    elif model_paths[0].suffix == ".pt":
+        # Use PyTorch generator (.pt)
+        if len(model_paths) > 1:
+            _LOGGER.error("Only one generator (.pt) is supported")
+            return 1
+
+        args["model"] = args["model"][0]
+        generate_samples(**args)
+    else:
+        _LOGGER.error("Models must have .pt or .onnx suffix")
+        return 1
+
+    return 0
+
+
+if __name__ == "__main__":
+    main()

piper_sample_generator/augment.py

@@ -1,12 +1,11 @@
 #!/usr/bin/env python3
 import argparse
 import audioop
-import sys
 import wave
 from pathlib import Path
 
 import numpy as np
-from audiomentations import Compose, ApplyImpulseResponse, Gain
+from audiomentations import ApplyImpulseResponse, Compose, Gain
 
 _DIR = Path(__file__).parent
 
@@ -15,14 +14,14 @@ def main() -> None:
     parser = argparse.ArgumentParser()
     parser.add_argument("input_dir")
     parser.add_argument("output_dir")
-    parser.add_argument("--sample-rate", type=int)
+    parser.add_argument("--sample-rate", type=int, required=True)
     args = parser.parse_args()
 
     impulses = list((_DIR / "impulses").glob("*.wav"))
 
     augment = Compose(
         transforms=[
-            Gain(min_gain_in_db=-12, max_gain_in_db=0),
+            Gain(min_gain_db=-12, max_gain_db=0),
             ApplyImpulseResponse(impulses),
         ]
     )
@@ -35,9 +34,10 @@ def main() -> None:
         output_wav = output_dir / (input_wav.relative_to(input_dir))
         output_wav.parent.mkdir(parents=True, exist_ok=True)
 
-        with wave.open(str(input_wav), "rb") as input_wav_file, wave.open(
+        with (
-            str(output_wav), "wb"
+            wave.open(str(input_wav), "rb") as input_wav_file,
-        ) as output_wav_file:
+            wave.open(str(output_wav), "wb") as output_wav_file,
+        ):
             assert input_wav_file.getsampwidth() == 2
             assert input_wav_file.getnchannels() == 1
 

pylintrc

@@ -1,3 +1,6 @@
+[MASTER]
+ignored-modules=torch
+
 [MESSAGES CONTROL]
 disable=
   format,
@@ -31,7 +34,7 @@ disable=
   missing-class-docstring,
   missing-function-docstring,
   import-error,
-  consider-using-with
+  relative-beyond-top-level
 
 [FORMAT]
 expected-line-ending-format=LF

pyproject.toml

@@ -0,0 +1,45 @@
+[build-system]
+requires = ["setuptools>=62.3"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name        = "piper-sample-generator"
+version     = "3.2.0"
+license     = {text = "MIT"}
+description = "Generate TTS audio samples for training wake word systems"
+readme      = "README.md"
+authors     = [
+    {name = "The Home Assistant Authors", email = "hello@home-assistant.io"}
+]
+keywords    = ["piper", "sample", "tts", "wakeword"]
+requires-python = ">=3.9.0"
+dependencies = [
+    "audiomentations==0.33.0",
+    "piper-tts==1.3.0",
+    "numpy>=2,<3",
+    "torch>=2,<3",
+    "torchaudio",
+    "webrtcvad",
+]
+
+[project.optional-dependencies]
+dev = [
+    "black==22.12.0",
+    "flake8==6.0.0",
+    "isort==5.11.3",
+    "mypy==0.991",
+    "pylint==2.15.9",
+]
+
+[project.urls]
+"Source Code" = "http://github.com/rhasspy/piper-sample-generator"
+
+[tool.setuptools]
+platforms = ["any"]
+zip-safe  = true
+
+[tool.setuptools.packages.find]
+include = ["piper_sample_generator*"]
+
+[tool.setuptools.package-data]
+piper_sample_generator = ["impulses/*.wav"]

requirements.txt

@@ -1,6 +0,0 @@
-audiomentations==0.33.0
-piper-phonemize==1.1.0
-numpy<2
-torch
-torchaudio
-webrtcvad

requirements_dev.txt

@@ -1,5 +0,0 @@
-black==22.12.0
-flake8==6.0.0
-isort==5.11.3
-mypy==0.991
-pylint==2.15.9

script/format

@@ -6,8 +6,13 @@ from pathlib import Path
 _DIR = Path(__file__).parent
 _PROGRAM_DIR = _DIR.parent
 _VENV_DIR = _PROGRAM_DIR / ".venv"
-_SCRIPT = _PROGRAM_DIR / "generate_samples.py"
+_MODULE_DIR = _PROGRAM_DIR / "piper_sample_generator"
 
-context = venv.EnvBuilder().ensure_directories(_VENV_DIR)
+if _VENV_DIR.exists():
-subprocess.check_call([context.env_exe, "-m", "black", str(_SCRIPT)])
+    context = venv.EnvBuilder().ensure_directories(_VENV_DIR)
-subprocess.check_call([context.env_exe, "-m", "isort", str(_SCRIPT)])
+    python_exe = context.env_exe
+else:
+    python_exe = "python3"
+
+subprocess.check_call([python_exe, "-m", "black", str(_MODULE_DIR)])
+subprocess.check_call([python_exe, "-m", "isort", str(_MODULE_DIR)])

script/lint

@@ -6,11 +6,16 @@ from pathlib import Path
 _DIR = Path(__file__).parent
 _PROGRAM_DIR = _DIR.parent
 _VENV_DIR = _PROGRAM_DIR / ".venv"
-_SCRIPT = _PROGRAM_DIR / "generate_samples.py"
+_MODULE_DIR = _PROGRAM_DIR / "piper_sample_generator"
 
-context = venv.EnvBuilder().ensure_directories(_VENV_DIR)
+if _VENV_DIR.exists():
-subprocess.check_call([context.env_exe, "-m", "black", str(_SCRIPT), "--check"])
+    context = venv.EnvBuilder().ensure_directories(_VENV_DIR)
-subprocess.check_call([context.env_exe, "-m", "isort", str(_SCRIPT), "--check"])
+    python_exe = context.env_exe
-subprocess.check_call([context.env_exe, "-m", "flake8", str(_SCRIPT)])
+else:
-subprocess.check_call([context.env_exe, "-m", "pylint", str(_SCRIPT)])
+    python_exe = "python3"
-subprocess.check_call([context.env_exe, "-m", "mypy", str(_SCRIPT)])
+
+subprocess.check_call([python_exe, "-m", "black", str(_MODULE_DIR), "--check"])
+subprocess.check_call([python_exe, "-m", "isort", str(_MODULE_DIR), "--check"])
+subprocess.check_call([python_exe, "-m", "flake8", str(_MODULE_DIR)])
+subprocess.check_call([python_exe, "-m", "pylint", str(_MODULE_DIR)])
+subprocess.check_call([python_exe, "-m", "mypy", str(_MODULE_DIR)])

script/run

@@ -8,5 +8,10 @@ _DIR = Path(__file__).parent
 _PROGRAM_DIR = _DIR.parent
 _VENV_DIR = _PROGRAM_DIR / ".venv"
 
-context = venv.EnvBuilder().ensure_directories(_VENV_DIR)
+if _VENV_DIR.exists():
-subprocess.check_call([context.env_exe, "generate_samples.py"] + sys.argv[1:])
+    context = venv.EnvBuilder().ensure_directories(_VENV_DIR)
+    python_exe = context.env_exe
+else:
+    python_exe = "python3"
+
+subprocess.check_call([python_exe, "-m", "piper_sample_generator"] + sys.argv[1:])

script/setup

@@ -1,4 +1,5 @@
 #!/usr/bin/env python3
+import argparse
 import subprocess
 import venv
 from pathlib import Path
@@ -7,6 +8,9 @@ _DIR = Path(__file__).parent
 _PROGRAM_DIR = _DIR.parent
 _VENV_DIR = _PROGRAM_DIR / ".venv"
 
+parser = argparse.ArgumentParser()
+parser.add_argument("--dev", action="store_true", help="Install dev requirements")
+args = parser.parse_args()
 
 # Create virtual environment
 builder = venv.EnvBuilder(with_pip=True)
@@ -19,4 +23,10 @@ subprocess.check_call(pip + ["install", "--upgrade", "pip"])
 subprocess.check_call(pip + ["install", "--upgrade", "setuptools", "wheel"])
 
 # Install requirements
-subprocess.check_call(pip + ["install", "-r", str(_PROGRAM_DIR / "requirements.txt")])
+subprocess.check_call(pip + ["install", "-e", str(_PROGRAM_DIR)])
+
+if args.dev:
+    # Install dev requirements
+    subprocess.check_call(
+        pip + ["install", "-e", f"{_PROGRAM_DIR}[dev]"]
+    )