run.py

import dataclasses
import json
import logging
from argparse import ArgumentParser
from pathlib import Path
from typing import Tuple

from voicevox_core import AccelerationMode, AudioQuery, wav_from_s16le
from voicevox_core.blocking import Onnxruntime, OpenJtalk, Synthesizer, VoiceModelFile


def main() -> None:
    logging.basicConfig(format="[%(levelname)s] %(name)s: %(message)s")
    logger = logging.getLogger(__name__)
    logger.setLevel("DEBUG")
    logging.getLogger("voicevox_core_python_api").setLevel("DEBUG")
    logging.getLogger("voicevox_core").setLevel("DEBUG")

    (
        acceleration_mode,
        vvm_path,
        onnxruntime_filename,
        open_jtalk_dict_dir,
        text,
        out,
        style_id,
        streaming,
    ) = parse_args()

    logger.info("%s", f"Loading ONNX Runtime ({onnxruntime_filename=})")
    onnxruntime = Onnxruntime.load_once(filename=onnxruntime_filename)

    logger.debug("%s", f"{onnxruntime.supported_devices()=}")

    logger.info("%s", f"Initializing ({acceleration_mode=}, {open_jtalk_dict_dir=})")
    synthesizer = Synthesizer(
        onnxruntime,
        OpenJtalk(open_jtalk_dict_dir),
        acceleration_mode=acceleration_mode,
    )

    logger.debug("%s", f"{synthesizer.metas=}")
    logger.debug("%s", f"{synthesizer.is_gpu_mode=}")

    logger.info("%s", f"Loading `{vvm_path}`")
    with VoiceModelFile.open(vvm_path) as model:
        synthesizer.load_voice_model(model)

    logger.info("%s", f"Creating an AudioQuery from {text!r}")
    audio_query = synthesizer.audio_query(text, style_id)

    logger.info("%s", f"Synthesizing with {display_as_json(audio_query)}")
    if streaming:
        logger.info("%s", "In streaming mode")
        chunk_sec = 1.0
        intermediate = synthesizer.precompute_render(audio_query, style_id)
        chunk_frames = int(intermediate.frame_rate * chunk_sec)
        pcm = b""
        for i in range(0, intermediate.frame_length, chunk_frames):
            logger.info("%s", f"{i/intermediate.frame_length:.2%}")
            pcm += synthesizer.render(intermediate, i, i + chunk_frames)
        logger.info("%s", f"100%")
        wav = wav_from_s16le(
            pcm, audio_query.output_sampling_rate, audio_query.output_stereo
        )

    else:
        wav = synthesizer.synthesis(audio_query, style_id)

    out.write_bytes(wav)
    logger.info("%s", f"Wrote `{out}`")


def parse_args() -> Tuple[AccelerationMode, Path, str, Path, str, Path, int, bool]:
    argparser = ArgumentParser()
    argparser.add_argument(
        "--mode",
        default="AUTO",
        type=AccelerationMode,
        help='モード ("AUTO", "CPU", "GPU")',
    )
    argparser.add_argument(
        "vvm",
        type=Path,
        help="vvmファイルへのパス",
    )
    argparser.add_argument(
        "--onnxruntime",
        default=Onnxruntime.LIB_VERSIONED_FILENAME,
        help="ONNX Runtimeのライブラリのfilename",
    )
    argparser.add_argument(
        "--dict-dir",
        default="./open_jtalk_dic_utf_8-1.11",
        type=Path,
        help="Open JTalkの辞書ディレクトリ",
    )
    argparser.add_argument(
        "--text",
        default="この音声は、ボイスボックスを使用して、出力されています。",
        help="読み上げさせたい文章",
    )
    argparser.add_argument(
        "--out",
        default="./output.wav",
        type=Path,
        help="出力wavファイルのパス",
    )
    argparser.add_argument(
        "--style-id",
        default=0,
        type=int,
        help="話者IDを指定",
    )
    argparser.add_argument(
        "--streaming",
        action="store_true",
        help="ストリーミング生成",
    )
    args = argparser.parse_args()
    # FIXME: 流石に多くなってきたので、`dataclass`化する
    return (
        args.mode,
        args.vvm,
        args.onnxruntime,
        args.dict_dir,
        args.text,
        args.out,
        args.style_id,
        args.streaming,
    )


def display_as_json(audio_query: AudioQuery) -> str:
    return json.dumps(dataclasses.asdict(audio_query), ensure_ascii=False)


if __name__ == "__main__":
    main()