gen_embeds.py

import argparse
from ctypes import alignment
import os
import sys
sys.path.append('rtvc/')
from pathlib import Path
import spacy
import matplotlib.pyplot as plt

import librosa
import numpy as np
import soundfile as sf
import torch
import noisereduce as nr  

from rtvc.encoder import inference as encoder
from rtvc.encoder.params_data import *
from rtvc.synthesizer.inference import Synthesizer_infer
from rtvc.utils.argutils import print_args
from rtvc.utils.default_models import ensure_default_models
from rtvc.vocoder import inference as vocoder
from rtvc.speed_changer.fixSpeed import *


if __name__ == '__main__':
    parser = argparse.ArgumentParser(
        formatter_class=argparse.ArgumentDefaultsHelpFormatter
    )
    parser.add_argument("--run_id", type=str, default="default", help= \
    "Name for this model. By default, training outputs will be stored to saved_models/<run_id>/. If a model state "
    "from the same run ID was previously saved, the training will restart from there. Pass -f to overwrite saved "
    "states and restart from scratch.")
    parser.add_argument("-m", "--models_dir", type=Path, default="rtvc/saved_models",
                        help="Directory containing all saved models")
    parser.add_argument("--weight", type=float, default=1,
                        help="weight of input audio for voice filter")
    parser.add_argument("--griffin_lim",
                        action="store_true",
                        help="if True, use vocoder, else use griffin-lim")
    parser.add_argument("--cpu", action="store_true", help=\
        "If True, processing is done on CPU, even when a GPU is available.")
    parser.add_argument("--no_sound", action="store_true", help=\
        "If True, audio won't be played.")
    parser.add_argument("--seed", type=int, default=None, help=\
        "Optional random number seed value to make toolbox deterministic.")
    args = parser.parse_args()
    arg_dict = vars(args)
    print_args(args, parser)

    # Hide GPUs from Pytorch to force CPU processing
    if arg_dict.pop("cpu"):
        os.environ["CUDA_VISIBLE_DEVICES"] = "-1"

    print("Running a test of your configuration...\n")

    if torch.cuda.is_available():
        device_id = torch.cuda.current_device()
        gpu_properties = torch.cuda.get_device_properties(device_id)
        ## Print some environment information (for debugging purposes)
        print("Found %d GPUs available. Using GPU %d (%s) of compute capability %d.%d with "
            "%.1fGb total memory.\n" %
            (torch.cuda.device_count(),
            device_id,
            gpu_properties.name,
            gpu_properties.major,
            gpu_properties.minor,
            gpu_properties.total_memory / 1e9))
    else:
        print("Using CPU for inference.\n")

    ## Load the models one by one.
    if not args.griffin_lim:
        print("Preparing the encoder, the synthesizer and the vocoder...")
    else:
        print("Preparing the encoder and the synthesizer...")
    ensure_default_models(args.run_id, Path("rtvc/saved_models"))
    encoder.load_model(list(args.models_dir.glob(f"{args.run_id}/encoder.pt"))[0])
    synthesizer = Synthesizer_infer(list(args.models_dir.glob(f"{args.run_id}/synthesizer.pt"))[0])
    if not args.griffin_lim:
        vocoder.load_model(list(args.models_dir.glob(f"{args.run_id}/vocoder.pt"))[0])

    nlp = spacy.load('en_core_web_sm')
    weight = arg_dict["weight"] # 声音美颜的用户语音权重
    amp = 1

    directory = "input_audios"
    pathlist = Path(directory).rglob('*.*')
    for path in pathlist:
        path = str(path)
        print(path)
        # enter the number of reference audios

        # Computing the embedding
        # First, we load the wav using the function that the speaker encoder provides. This is
        # important: there is preprocessing that must be applied.

        # The following two methods are equivalent:
        # - Directly load from the filepath:
        # preprocessed_wav = encoder.preprocess_wav(in_fpath)
        # - If the wav is already loaded:

        # get duration info from input audio
        in_fpath = Path(path.replace("\"", "").replace("\'", ""))
        fpath_without_ext = os.path.splitext(str(in_fpath))[0]
        speaker_name = os.path.normpath(fpath_without_ext).split(os.sep)[-1]

        is_wav_file, wav, wav_path = TransFormat(in_fpath, 'wav')
        # 除了m4a格式无法工作而必须转换以外，无论原格式是否为wav，从稳定性的角度考虑也最好再转为wav（因为某些wav本身不带比特率属性，无法在此代码中工作，因此需要转换以赋予其该属性）

        if not is_wav_file:
            os.remove(wav_path)  # remove intermediate wav files

        preprocessed_wav = encoder.preprocess_wav(wav)

        print("Loaded input audio file succesfully")

        # Then we derive the embedding. There are many functions and parameters that the
        # speaker encoder interfaces. These are mostly for in-depth research. You will typically
        # only use this function (with its default parameters):
        embed = encoder.embed_utterance(preprocessed_wav)
        embed[embed < set_zero_thres]=0 # 噪声值置零
        if not os.path.exists("embeds"):
            os.mkdir("embeds")
        np.save(f"embeds/{speaker_name}.npy", embed)