VoxCPM/inference.py

import soundfile as sf
import numpy as np
from voxcpm import VoxCPM
import argparse
import os
def main():
    parser = argparse.ArgumentParser()
    parser.add_argument("--model_path", type=str, required=True)
    parser.add_argument("--text", type=str)
    parser.add_argument("--text_file", type=str)
    parser.add_argument("--output_dir", type=str, default="outputs")
    parser.add_argument("--cfg_value", type=float, default=2.0)
    parser.add_argument("--inference_timesteps", type=int, default=10)
    parser.add_argument("--prompt_wav_path", type=str)
    parser.add_argument("--prompt_text", type=str)
    args = parser.parse_args()
    assert args.text or args.text_file, "Please provide either text or text_file"
    # validate prompt_wav_path and prompt_text 必须同时提供
    if args.prompt_wav_path or args.prompt_text:
        assert args.prompt_wav_path and args.prompt_text, "Please provide both prompt_wav_path and prompt_text"
    model = VoxCPM.from_pretrained(args.model_path, load_denoiser=False)
    if args.text:
        wav = model.generate(
        text=args.text,
        prompt_wav_path=args.prompt_wav_path,      # optional: path to a prompt speech for voice cloning
        prompt_text=args.prompt_text,          # optional: reference text
        cfg_value=args.cfg_value,             # LM guidance on LocDiT, higher for better adherence to the prompt, but maybe worse
        inference_timesteps=args.inference_timesteps,   # LocDiT inference timesteps, higher for better result, lower for fast speed
        normalize=True,           # enable external TN tool
        denoise=False,             # enable external Denoise tool
        retry_badcase=True,        # enable retrying mode for some bad cases (unstoppable)
        retry_badcase_max_times=3,  # maximum retrying times
        retry_badcase_ratio_threshold=6.0, # maximum length restriction for bad case detection (simple but effective), it could be adjusted for slow pace speech
        )
        if not os.path.exists(args.output_dir):
            os.makedirs(args.output_dir)
        sf.write(f"{args.output_dir}/output.wav", wav, 16000)
        print(f"saved: {args.output_dir}/output.wav")
    elif args.text_file:
        texts = []
        with open(args.text_file, "r") as f:
            lines = f.readlines()
            for line in lines:
                line = line.strip().split("||")
                wav_id = line[0]
                text = " ".join(line[1:])
                texts.append((wav_id, text))
        for wav_id, text in texts:
            wav = model.generate(
            text=text,
            prompt_wav_path=args.prompt_wav_path,      # optional: path to a prompt speech for voice cloning
            prompt_text=args.prompt_text,          # optional: reference text
            cfg_value=args.cfg_value,             # LM guidance on LocDiT, higher for better adherence to the prompt, but maybe worse
            inference_timesteps=args.inference_timesteps,   # LocDiT inference timesteps, higher for better result, lower for fast speed
            normalize=True,           # enable external TN tool
            denoise=False,             # enable external Denoise tool
            retry_badcase=True,        # enable retrying mode for some bad cases (unstoppable)
            retry_badcase_max_times=3,  # maximum retrying times
            retry_badcase_ratio_threshold=6.0, # maximum length restriction for bad case detection (simple but effective), it could be adjusted for slow pace speech
            )
            if not os.path.exists(args.output_dir):
                os.makedirs(args.output_dir)
            sf.write(f"{args.output_dir}/{wav_id}.wav", wav, 16000)
            print(f"saved: {args.output_dir}/{wav_id}.wav")

if __name__ == "__main__":
    main()