vits-coqui.py

#!/usr/bin/env python3

import collections
import os
from typing import Any, Dict

import onnx
from TTS.tts.configs.vits_config import VitsConfig
from TTS.tts.models.vits import Vits


def add_meta_data(filename: str, meta_data: Dict[str, Any]):
    """Add meta data to an ONNX model. It is changed in-place.

    Args:
      filename:
        Filename of the ONNX model to be changed.
      meta_data:
        Key-value pairs.
    """
    model = onnx.load(filename)
    for key, value in meta_data.items():
        meta = model.metadata_props.add()
        meta.key = key
        meta.value = str(value)

    onnx.save(model, filename)


lang_map = {
    "bg": "Bulgarian",
    "bn": "Bangla",
    "cs": "Czech",
    "da": "Danish",
    "de": "German",
    "el": "Greek",
    "es": "Spanish",
    "et": "Estonian",
    "fr": "French",
    "ga": "Irish",
    "fi": "Finnish",
    "hr": "Croatian",
    "hu": "Hungarian",
    "is": "Icelandic",
    "it": "Italian",
    "ka": "Georgian",
    "kk": "Kazakh",
    "lb": "Luxembourgish",
    "lt": "Lithuanian",
    "lv": "Latvian",
    "mt": "Maltese",
    "ne": "Nepali",
    "nl": "Dutch",
    "no": "Norwegian",
    "pl": "Polish",
    "pt": "Portuguese",
    "ro": "Romanian",
    "sl": "Slovenian",
    "sk": "Slovak",
    "sr": "Serbian",
    "sv": "Swedish",
    "sw": "Swahili",
    "tr": "Turkish",
    "uk": "Ukrainian",
    "vi": "Vietnamese",
    "zh": "Chinese",
}


def main():
    lang = os.environ.get("LANG", None)
    if not lang:
        print("Please provide the environment variable LANG")
        return

    config = VitsConfig()
    config.load_json("config.json")

    # Initialize VITS model and load its checkpoint
    vits = Vits.init_from_config(config)

    assert vits.config.use_phonemes is False, vits.config.use_phonemes
    assert vits.config.phonemizer is None, vits.config.phonemizer
    assert vits.config.phoneme_language is None, vits.config.phoneme_language

    vits.load_checkpoint(config, "model_file.pth")
    vits.export_onnx(output_path="model.onnx", verbose=False)

    language = lang_map[lang]

    meta_data = {
        "model_type": "vits",
        "comment": "coqui",  # For models from coqui-ai/TTS, it must be coqui
        "language": language,
        "frontend": "characters",
        "add_blank": int(vits.config.add_blank),
        "blank_id": vits.tokenizer.characters.blank_id,
        "n_speakers": vits.config.model_args.num_speakers,
        "use_eos_bos": int(vits.tokenizer.use_eos_bos),
        "bos_id": vits.tokenizer.characters.bos_id,
        "eos_id": vits.tokenizer.characters.eos_id,
        "pad_id": vits.tokenizer.characters.pad_id,
        "sample_rate": int(vits.ap.sample_rate),
    }
    print("meta_data", meta_data)
    add_meta_data(filename="model.onnx", meta_data=meta_data)

    # Now generate tokens.txt
    all_upper_tokens = [i.upper() for i in vits.tokenizer.characters._char_to_id.keys()]
    duplicate = set(
        [
            item
            for item, count in collections.Counter(all_upper_tokens).items()
            if count > 1
        ]
    )

    with open("tokens.txt", "w", encoding="utf-8") as f:
        for token, idx in vits.tokenizer.characters._char_to_id.items():
            f.write(f"{token} {idx}\n")

            # both upper case and lower case correspond to the same ID
            if (
                token not in ("<PAD>", "<EOS>", "BOS", "<BLNK>")
                and token.lower() != token.upper()
                and len(token.upper()) == 1
                and token.upper() not in duplicate
            ):
                f.write(f"{token.upper()} {idx}\n")


if __name__ == "__main__":
    main()