Spaces:
Sleeping
Sleeping
File size: 4,872 Bytes
05005db |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 |
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
# Copyright 2019 Tomoki Hayashi
# MIT License (https://opensource.org/licenses/MIT)
# Modified by Yiwei Guo, 2024
"""Decode with trained vec2wav Generator."""
import argparse
import logging
import os
import time
import numpy as np
import soundfile as sf
import torch
import yaml
from tqdm import tqdm
from vec2wav2.datasets import MelSCPDataset
from vec2wav2.utils import load_model, load_feat_codebook, idx2vec
def set_loglevel(verbose):
# set logger
if verbose > 1:
logging.basicConfig(
level=logging.DEBUG,
format="%(asctime)s (%(module)s:%(lineno)d) %(levelname)s: %(message)s",
)
elif verbose > 0:
logging.basicConfig(
level=logging.INFO,
format="%(asctime)s (%(module)s:%(lineno)d) %(levelname)s: %(message)s",
)
else:
logging.basicConfig(
level=logging.WARN,
format="%(asctime)s (%(module)s:%(lineno)d) %(levelname)s: %(message)s",
)
logging.warning("Skip DEBUG/INFO messages")
def main():
"""Run decoding process."""
parser = argparse.ArgumentParser(
description="Decode from audio tokens and acoustic prompts with trained vec2wav model"
"(See detail in vec2wav2/bin/decode.py)."
)
parser.add_argument(
"--feats-scp",
"--scp",
default=None,
type=str,
required=True,
help="kaldi-style feats.scp file. "
)
parser.add_argument(
"--prompt-scp",
default=None,
type=str,
help="kaldi-style prompt.scp file. Similar to feats.scp."
)
parser.add_argument(
"--outdir",
type=str,
required=True,
help="directory to save generated speech.",
)
parser.add_argument(
"--checkpoint",
type=str,
required=True,
help="checkpoint file to be loaded.",
)
parser.add_argument(
"--config",
default=None,
type=str,
help="yaml format configuration file. if not explicitly provided, "
"it will be searched in the checkpoint directory. (default=None)",
)
parser.add_argument(
"--verbose",
type=int,
default=1,
help="logging level. higher is more logging. (default=1)",
)
args = parser.parse_args()
set_loglevel(args.verbose)
# check directory existence
if not os.path.exists(args.outdir):
os.makedirs(args.outdir)
# load config
if args.config is None:
dirname = os.path.dirname(args.checkpoint)
args.config = os.path.join(dirname, "config.yml")
with open(args.config) as f:
config = yaml.load(f, Loader=yaml.Loader)
config.update(vars(args))
# get dataset
dataset = MelSCPDataset(
vqidx_scp=args.feats_scp,
prompt_scp=args.prompt_scp,
return_utt_id=True,
)
logging.info(f"The number of features to be decoded = {len(dataset)}.")
# setup model
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
logging.info(f"Using {'GPU' if torch.cuda.is_available() else 'CPU'}.")
model = load_model(args.checkpoint, config)
logging.info(f"Loaded model parameters from {args.checkpoint}.")
model.backend.remove_weight_norm()
model = model.eval().to(device)
# load vq codebook
feat_codebook, feat_codebook_numgroups = load_feat_codebook(np.load(config["vq_codebook"], allow_pickle=True), device)
# start generation
total_rtf = 0.0
with torch.no_grad(), tqdm(dataset, desc="[decode]") as pbar:
for idx, batch in enumerate(pbar, 1):
utt_id, vqidx, prompt = batch[0], batch[1], batch[2]
vqidx = torch.tensor(vqidx).to(device) # (L, G)
prompt = torch.tensor(prompt).unsqueeze(0).to(device) # (1, L', D')
vqidx = vqidx.long()
vqvec = idx2vec(feat_codebook, vqidx, feat_codebook_numgroups).unsqueeze(0) # (1, L, D)
# generate
start = time.time()
y = model.inference(vqvec, prompt)[-1].view(-1)
rtf = (time.time() - start) / (len(y) / config["sampling_rate"])
pbar.set_postfix({"RTF": rtf})
total_rtf += rtf
tgt_dir = os.path.dirname(os.path.join(config["outdir"], f"{utt_id}.wav"))
os.makedirs(tgt_dir, exist_ok=True)
basename = os.path.basename(f"{utt_id}.wav")
# save as PCM 16 bit wav file
sf.write(
os.path.join(tgt_dir, basename),
y.cpu().numpy(),
config["sampling_rate"],
"PCM_16",
)
# report average RTF
logging.info(f"Finished generation of {idx} utterances (RTF = {total_rtf / idx:.03f}).")
if __name__ == "__main__":
main()
|