import re import os import gc import sys import time import faiss import torch import librosa import logging import argparse import warnings import onnxruntime import logging.handlers import numpy as np import soundfile as sf import torch.nn.functional as F from tqdm import tqdm from scipy import signal from distutils.util import strtobool warnings.filterwarnings("ignore") sys.path.append(os.getcwd()) from main.configs.config import Config from main.library.algorithm.synthesizers import Synthesizer from main.library.utils import check_predictors, check_embedders, load_audio, load_embedders_model, cut, restore bh, ah = signal.butter(N=5, Wn=48, btype="high", fs=16000) config = Config() translations = config.translations logger = logging.getLogger(__name__) logger.propagate = False for l in ["torch", "faiss", "httpx", "httpcore", "faiss.loader", "numba.core", "urllib3", "transformers", "matplotlib"]: logging.getLogger(l).setLevel(logging.ERROR) if logger.hasHandlers(): logger.handlers.clear() else: console_handler = logging.StreamHandler() console_formatter = logging.Formatter(fmt="\n%(asctime)s.%(msecs)03d | %(levelname)s | %(module)s | %(message)s", datefmt="%Y-%m-%d %H:%M:%S") console_handler.setFormatter(console_formatter) console_handler.setLevel(logging.INFO) file_handler = logging.handlers.RotatingFileHandler(os.path.join("assets", "logs", "convert.log"), maxBytes=5*1024*1024, backupCount=3, encoding='utf-8') file_formatter = logging.Formatter(fmt="\n%(asctime)s.%(msecs)03d | %(levelname)s | %(module)s | %(message)s", datefmt="%Y-%m-%d %H:%M:%S") file_handler.setFormatter(file_formatter) file_handler.setLevel(logging.DEBUG) logger.addHandler(console_handler) logger.addHandler(file_handler) logger.setLevel(logging.DEBUG) def parse_arguments(): parser = argparse.ArgumentParser() parser.add_argument("--pitch", type=int, default=0) parser.add_argument("--filter_radius", type=int, default=3) parser.add_argument("--index_rate", type=float, default=0.5) parser.add_argument("--volume_envelope", type=float, default=1) parser.add_argument("--protect", type=float, default=0.33) parser.add_argument("--hop_length", type=int, default=64) parser.add_argument("--f0_method", type=str, default="rmvpe") parser.add_argument("--embedder_model", type=str, default="contentvec_base") parser.add_argument("--input_path", type=str, required=True) parser.add_argument("--output_path", type=str, default="./audios/output.wav") parser.add_argument("--export_format", type=str, default="wav") parser.add_argument("--pth_path", type=str, required=True) parser.add_argument("--index_path", type=str) parser.add_argument("--f0_autotune", type=lambda x: bool(strtobool(x)), default=False) parser.add_argument("--f0_autotune_strength", type=float, default=1) parser.add_argument("--clean_audio", type=lambda x: bool(strtobool(x)), default=False) parser.add_argument("--clean_strength", type=float, default=0.7) parser.add_argument("--resample_sr", type=int, default=0) parser.add_argument("--split_audio", type=lambda x: bool(strtobool(x)), default=False) parser.add_argument("--checkpointing", type=lambda x: bool(strtobool(x)), default=False) parser.add_argument("--f0_file", type=str, default="") parser.add_argument("--f0_onnx", type=lambda x: bool(strtobool(x)), default=False) parser.add_argument("--embedders_mode", type=str, default="fairseq") parser.add_argument("--formant_shifting", type=lambda x: bool(strtobool(x)), default=False) parser.add_argument("--formant_qfrency", type=float, default=0.8) parser.add_argument("--formant_timbre", type=float, default=0.8) return parser.parse_args() def main(): args = parse_arguments() pitch, filter_radius, index_rate, volume_envelope, protect, hop_length, f0_method, input_path, output_path, pth_path, index_path, f0_autotune, f0_autotune_strength, clean_audio, clean_strength, export_format, embedder_model, resample_sr, split_audio, checkpointing, f0_file, f0_onnx, embedders_mode, formant_shifting, formant_qfrency, formant_timbre = args.pitch, args.filter_radius, args.index_rate, args.volume_envelope,args.protect, args.hop_length, args.f0_method, args.input_path, args.output_path, args.pth_path, args.index_path, args.f0_autotune, args.f0_autotune_strength, args.clean_audio, args.clean_strength, args.export_format, args.embedder_model, args.resample_sr, args.split_audio, args.checkpointing, args.f0_file, args.f0_onnx, args.embedders_mode, args.formant_shifting, args.formant_qfrency, args.formant_timbre log_data = {translations['pitch']: pitch, translations['filter_radius']: filter_radius, translations['index_strength']: index_rate, translations['volume_envelope']: volume_envelope, translations['protect']: protect, "Hop length": hop_length, translations['f0_method']: f0_method, translations['audio_path']: input_path, translations['output_path']: output_path.replace('wav', export_format), translations['model_path']: pth_path, translations['indexpath']: index_path, translations['autotune']: f0_autotune, translations['clear_audio']: clean_audio, translations['export_format']: export_format, translations['hubert_model']: embedder_model, translations['split_audio']: split_audio, translations['memory_efficient_training']: checkpointing, translations["f0_onnx_mode"]: f0_onnx, translations["embed_mode"]: embedders_mode} if clean_audio: log_data[translations['clean_strength']] = clean_strength if resample_sr != 0: log_data[translations['sample_rate']] = resample_sr if f0_autotune: log_data[translations['autotune_rate_info']] = f0_autotune_strength if os.path.isfile(f0_file): log_data[translations['f0_file']] = f0_file if formant_shifting: log_data[translations['formant_qfrency']] = formant_qfrency log_data[translations['formant_timbre']] = formant_timbre for key, value in log_data.items(): logger.debug(f"{key}: {value}") run_convert_script(pitch=pitch, filter_radius=filter_radius, index_rate=index_rate, volume_envelope=volume_envelope, protect=protect, hop_length=hop_length, f0_method=f0_method, input_path=input_path, output_path=output_path, pth_path=pth_path, index_path=index_path, f0_autotune=f0_autotune, f0_autotune_strength=f0_autotune_strength, clean_audio=clean_audio, clean_strength=clean_strength, export_format=export_format, embedder_model=embedder_model, resample_sr=resample_sr, split_audio=split_audio, checkpointing=checkpointing, f0_file=f0_file, f0_onnx=f0_onnx, embedders_mode=embedders_mode, formant_shifting=formant_shifting, formant_qfrency=formant_qfrency, formant_timbre=formant_timbre) def run_convert_script(pitch=0, filter_radius=3, index_rate=0.5, volume_envelope=1, protect=0.5, hop_length=64, f0_method="rmvpe", input_path=None, output_path="./output.wav", pth_path=None, index_path=None, f0_autotune=False, f0_autotune_strength=1, clean_audio=False, clean_strength=0.7, export_format="wav", embedder_model="contentvec_base", resample_sr=0, split_audio=False, checkpointing=False, f0_file=None, f0_onnx=False, embedders_mode="fairseq", formant_shifting=False, formant_qfrency=0.8, formant_timbre=0.8): check_predictors(f0_method, f0_onnx); check_embedders(embedder_model, embedders_mode) if not pth_path or not os.path.exists(pth_path) or os.path.isdir(pth_path) or not pth_path.endswith((".pth", ".onnx")): logger.warning(translations["provide_file"].format(filename=translations["model"])) sys.exit(1) cvt = VoiceConverter(pth_path, 0) start_time = time.time() pid_path = os.path.join("assets", "convert_pid.txt") with open(pid_path, "w") as pid_file: pid_file.write(str(os.getpid())) if os.path.isdir(input_path): logger.info(translations["convert_batch"]) audio_files = [f for f in os.listdir(input_path) if f.lower().endswith(("wav", "mp3", "flac", "ogg", "opus", "m4a", "mp4", "aac", "alac", "wma", "aiff", "webm", "ac3"))] if not audio_files: logger.warning(translations["not_found_audio"]) sys.exit(1) logger.info(translations["found_audio"].format(audio_files=len(audio_files))) for audio in audio_files: audio_path = os.path.join(input_path, audio) output_audio = os.path.join(input_path, os.path.splitext(audio)[0] + f"_output.{export_format}") logger.info(f"{translations['convert_audio']} '{audio_path}'...") if os.path.exists(output_audio): os.remove(output_audio) cvt.convert_audio(pitch=pitch, filter_radius=filter_radius, index_rate=index_rate, volume_envelope=volume_envelope, protect=protect, hop_length=hop_length, f0_method=f0_method, audio_input_path=audio_path, audio_output_path=output_audio, index_path=index_path, f0_autotune=f0_autotune, f0_autotune_strength=f0_autotune_strength, clean_audio=clean_audio, clean_strength=clean_strength, export_format=export_format, embedder_model=embedder_model, resample_sr=resample_sr, checkpointing=checkpointing, f0_file=f0_file, f0_onnx=f0_onnx, embedders_mode=embedders_mode, formant_shifting=formant_shifting, formant_qfrency=formant_qfrency, formant_timbre=formant_timbre, split_audio=split_audio) logger.info(translations["convert_batch_success"].format(elapsed_time=f"{(time.time() - start_time):.2f}", output_path=output_path.replace('wav', export_format))) else: if not os.path.exists(input_path): logger.warning(translations["not_found_audio"]) sys.exit(1) logger.info(f"{translations['convert_audio']} '{input_path}'...") if os.path.exists(output_path): os.remove(output_path) cvt.convert_audio(pitch=pitch, filter_radius=filter_radius, index_rate=index_rate, volume_envelope=volume_envelope, protect=protect, hop_length=hop_length, f0_method=f0_method, audio_input_path=input_path, audio_output_path=output_path, index_path=index_path, f0_autotune=f0_autotune, f0_autotune_strength=f0_autotune_strength, clean_audio=clean_audio, clean_strength=clean_strength, export_format=export_format, embedder_model=embedder_model, resample_sr=resample_sr, checkpointing=checkpointing, f0_file=f0_file, f0_onnx=f0_onnx, embedders_mode=embedders_mode, formant_shifting=formant_shifting, formant_qfrency=formant_qfrency, formant_timbre=formant_timbre, split_audio=split_audio) if os.path.exists(pid_path): os.remove(pid_path) logger.info(translations["convert_audio_success"].format(input_path=input_path, elapsed_time=f"{(time.time() - start_time):.2f}", output_path=output_path.replace('wav', export_format))) def change_rms(source_audio, source_rate, target_audio, target_rate, rate): rms2 = F.interpolate(torch.from_numpy(librosa.feature.rms(y=target_audio, frame_length=target_rate // 2 * 2, hop_length=target_rate // 2)).float().unsqueeze(0), size=target_audio.shape[0], mode="linear").squeeze() return (target_audio * (torch.pow(F.interpolate(torch.from_numpy(librosa.feature.rms(y=source_audio, frame_length=source_rate // 2 * 2, hop_length=source_rate // 2)).float().unsqueeze(0), size=target_audio.shape[0], mode="linear").squeeze(), 1 - rate) * torch.pow(torch.maximum(rms2, torch.zeros_like(rms2) + 1e-6), rate - 1)).numpy()) def clear_gpu_cache(): gc.collect() if torch.cuda.is_available(): torch.cuda.empty_cache() elif torch.backends.mps.is_available(): torch.mps.empty_cache() def get_providers(): ort_providers = onnxruntime.get_available_providers() if "CUDAExecutionProvider" in ort_providers: providers = ["CUDAExecutionProvider"] elif "CoreMLExecutionProvider" in ort_providers: providers = ["CoreMLExecutionProvider"] else: providers = ["CPUExecutionProvider"] return providers class Autotune: def __init__(self, ref_freqs): self.ref_freqs = ref_freqs self.note_dict = self.ref_freqs def autotune_f0(self, f0, f0_autotune_strength): autotuned_f0 = np.zeros_like(f0) for i, freq in enumerate(f0): autotuned_f0[i] = freq + (min(self.note_dict, key=lambda x: abs(x - freq)) - freq) * f0_autotune_strength return autotuned_f0 class VC: def __init__(self, tgt_sr, config): self.x_pad = config.x_pad self.x_query = config.x_query self.x_center = config.x_center self.x_max = config.x_max self.sample_rate = 16000 self.window = 160 self.t_pad = self.sample_rate * self.x_pad self.t_pad_tgt = tgt_sr * self.x_pad self.t_pad2 = self.t_pad * 2 self.t_query = self.sample_rate * self.x_query self.t_center = self.sample_rate * self.x_center self.t_max = self.sample_rate * self.x_max self.time_step = self.window / self.sample_rate * 1000 self.f0_min = 50 self.f0_max = 1100 self.f0_mel_min = 1127 * np.log(1 + self.f0_min / 700) self.f0_mel_max = 1127 * np.log(1 + self.f0_max / 700) self.device = config.device self.is_half = config.is_half self.ref_freqs = [49.00, 51.91, 55.00, 58.27, 61.74, 65.41, 69.30, 73.42, 77.78, 82.41, 87.31, 92.50, 98.00, 103.83, 110.00, 116.54, 123.47, 130.81, 138.59, 146.83, 155.56, 164.81, 174.61, 185.00, 196.00, 207.65, 220.00, 233.08, 246.94, 261.63, 277.18, 293.66, 311.13, 329.63, 349.23, 369.99, 392.00, 415.30, 440.00, 466.16, 493.88, 523.25, 554.37, 587.33, 622.25, 659.25, 698.46, 739.99, 783.99, 830.61, 880.00, 932.33, 987.77, 1046.50] self.autotune = Autotune(self.ref_freqs) self.note_dict = self.autotune.note_dict def get_f0_pm(self, x, p_len): import parselmouth f0 = (parselmouth.Sound(x, self.sample_rate).to_pitch_ac(time_step=self.window / self.sample_rate * 1000 / 1000, voicing_threshold=0.6, pitch_floor=self.f0_min, pitch_ceiling=self.f0_max).selected_array["frequency"]) pad_size = (p_len - len(f0) + 1) // 2 if pad_size > 0 or p_len - len(f0) - pad_size > 0: f0 = np.pad(f0, [[pad_size, p_len - len(f0) - pad_size]], mode="constant") return f0 def get_f0_mangio_crepe(self, x, p_len, hop_length, model="full", onnx=False): from main.library.predictors.CREPE import predict x = x.astype(np.float32) x /= np.quantile(np.abs(x), 0.999) audio = torch.unsqueeze(torch.from_numpy(x).to(self.device, copy=True), dim=0) if audio.ndim == 2 and audio.shape[0] > 1: audio = torch.mean(audio, dim=0, keepdim=True).detach() p_len = p_len or x.shape[0] // hop_length source = np.array(predict(audio.detach(), self.sample_rate, hop_length, self.f0_min, self.f0_max, model, batch_size=hop_length * 2, device=self.device, pad=True, providers=get_providers(), onnx=onnx).squeeze(0).cpu().float().numpy()) source[source < 0.001] = np.nan return np.nan_to_num(np.interp(np.arange(0, len(source) * p_len, len(source)) / p_len, np.arange(0, len(source)), source)) def get_f0_crepe(self, x, model="full", onnx=False): from main.library.predictors.CREPE import predict, mean, median f0, pd = predict(torch.tensor(np.copy(x))[None].float(), self.sample_rate, self.window, self.f0_min, self.f0_max, model, batch_size=512, device=self.device, return_periodicity=True, providers=get_providers(), onnx=onnx) f0, pd = mean(f0, 3), median(pd, 3) f0[pd < 0.1] = 0 return f0[0].cpu().numpy() def get_f0_fcpe(self, x, p_len, hop_length, onnx=False, legacy=False): from main.library.predictors.FCPE import FCPE model_fcpe = FCPE(os.path.join("assets", "models", "predictors", ("fcpe_legacy" if legacy else "fcpe") + (".onnx" if onnx else ".pt")), hop_length=int(hop_length), f0_min=int(self.f0_min), f0_max=int(self.f0_max), dtype=torch.float32, device=self.device, sample_rate=self.sample_rate, threshold=0.03 if legacy else 0.006, providers=get_providers(), onnx=onnx, legacy=legacy) f0 = model_fcpe.compute_f0(x, p_len=p_len) del model_fcpe return f0 def get_f0_rmvpe(self, x, legacy=False, onnx=False): from main.library.predictors.RMVPE import RMVPE rmvpe_model = RMVPE(os.path.join("assets", "models", "predictors", "rmvpe" + (".onnx" if onnx else ".pt")), is_half=self.is_half, device=self.device, onnx=onnx, providers=get_providers()) f0 = rmvpe_model.infer_from_audio_with_pitch(x, thred=0.03, f0_min=self.f0_min, f0_max=self.f0_max) if legacy else rmvpe_model.infer_from_audio(x, thred=0.03) del rmvpe_model return f0 def get_f0_pyworld(self, x, filter_radius, model="harvest"): from main.library.predictors.WORLD_WRAPPER import PYWORLD pw = PYWORLD() x = x.astype(np.double) if model == "harvest": f0, t = pw.harvest(x, fs=self.sample_rate, f0_ceil=self.f0_max, f0_floor=self.f0_min, frame_period=10) elif model == "dio": f0, t = pw.dio(x, fs=self.sample_rate, f0_ceil=self.f0_max, f0_floor=self.f0_min, frame_period=10) else: raise ValueError(translations["method_not_valid"]) f0 = pw.stonemask(x, self.sample_rate, t, f0) if filter_radius > 2 or model == "dio": f0 = signal.medfilt(f0, filter_radius) return f0 def get_f0_swipe(self, x): from main.library.predictors.SWIPE import swipe f0, _ = swipe(x.astype(np.float32), self.sample_rate, f0_floor=self.f0_min, f0_ceil=self.f0_max, frame_period=10) return f0 def get_f0_yin(self, x, hop_length, p_len, mode="yin"): source = np.array(librosa.yin(x.astype(np.float32), sr=self.sample_rate, fmin=self.f0_min, fmax=self.f0_max, hop_length=hop_length) if mode == "yin" else librosa.pyin(x.astype(np.float32), fmin=self.f0_min, fmax=self.f0_max, sr=self.sample_rate, hop_length=hop_length)[0]) source[source < 0.001] = np.nan return np.nan_to_num(np.interp(np.arange(0, len(source) * p_len, len(source)) / p_len, np.arange(0, len(source)), source)) def get_f0_hybrid(self, methods_str, x, p_len, hop_length, filter_radius, onnx_mode): methods_str = re.search("hybrid\[(.+)\]", methods_str) if methods_str: methods = [method.strip() for method in methods_str.group(1).split("+")] f0_computation_stack, resampled_stack = [], [] logger.debug(translations["hybrid_methods"].format(methods=methods)) x = x.astype(np.float32) x /= np.quantile(np.abs(x), 0.999) for method in methods: f0 = None f0_methods = {"pm": lambda: self.get_f0_pm(x, p_len), "dio": lambda: self.get_f0_pyworld(x, filter_radius, "dio"), "mangio-crepe-tiny": lambda: self.get_f0_mangio_crepe(x, p_len, int(hop_length), "tiny", onnx=onnx_mode), "mangio-crepe-small": lambda: self.get_f0_mangio_crepe(x, p_len, int(hop_length), "small", onnx=onnx_mode), "mangio-crepe-medium": lambda: self.get_f0_mangio_crepe(x, p_len, int(hop_length), "medium", onnx=onnx_mode), "mangio-crepe-large": lambda: self.get_f0_mangio_crepe(x, p_len, int(hop_length), "large", onnx=onnx_mode), "mangio-crepe-full": lambda: self.get_f0_mangio_crepe(x, p_len, int(hop_length), "full", onnx=onnx_mode), "crepe-tiny": lambda: self.get_f0_crepe(x, "tiny", onnx=onnx_mode), "crepe-small": lambda: self.get_f0_crepe(x, "small", onnx=onnx_mode), "crepe-medium": lambda: self.get_f0_crepe(x, "medium", onnx=onnx_mode), "crepe-large": lambda: self.get_f0_crepe(x, "large", onnx=onnx_mode), "crepe-full": lambda: self.get_f0_crepe(x, "full", onnx=onnx_mode), "fcpe": lambda: self.get_f0_fcpe(x, p_len, int(hop_length), onnx=onnx_mode), "fcpe-legacy": lambda: self.get_f0_fcpe(x, p_len, int(hop_length), legacy=True, onnx=onnx_mode), "rmvpe": lambda: self.get_f0_rmvpe(x, onnx=onnx_mode), "rmvpe-legacy": lambda: self.get_f0_rmvpe(x, legacy=True, onnx=onnx_mode), "harvest": lambda: self.get_f0_pyworld(x, filter_radius, "harvest"), "yin": lambda: self.get_f0_yin(x, int(hop_length), p_len, mode="yin"), "pyin": lambda: self.get_f0_yin(x, int(hop_length), p_len, mode="pyin"), "swipe": lambda: self.get_f0_swipe(x)} f0 = f0_methods.get(method, lambda: ValueError(translations["method_not_valid"]))() f0_computation_stack.append(f0) for f0 in f0_computation_stack: resampled_stack.append(np.interp(np.linspace(0, len(f0), p_len), np.arange(len(f0)), f0)) return resampled_stack[0] if len(resampled_stack) == 1 else np.nanmedian(np.vstack(resampled_stack), axis=0) def get_f0(self, x, p_len, pitch, f0_method, filter_radius, hop_length, f0_autotune, f0_autotune_strength, inp_f0=None, onnx_mode=False): f0_methods = {"pm": lambda: self.get_f0_pm(x, p_len), "dio": lambda: self.get_f0_pyworld(x, filter_radius, "dio"), "mangio-crepe-tiny": lambda: self.get_f0_mangio_crepe(x, p_len, int(hop_length), "tiny", onnx=onnx_mode), "mangio-crepe-small": lambda: self.get_f0_mangio_crepe(x, p_len, int(hop_length), "small", onnx=onnx_mode), "mangio-crepe-medium": lambda: self.get_f0_mangio_crepe(x, p_len, int(hop_length), "medium", onnx=onnx_mode), "mangio-crepe-large": lambda: self.get_f0_mangio_crepe(x, p_len, int(hop_length), "large", onnx=onnx_mode), "mangio-crepe-full": lambda: self.get_f0_mangio_crepe(x, p_len, int(hop_length), "full", onnx=onnx_mode), "crepe-tiny": lambda: self.get_f0_crepe(x, "tiny", onnx=onnx_mode), "crepe-small": lambda: self.get_f0_crepe(x, "small", onnx=onnx_mode), "crepe-medium": lambda: self.get_f0_crepe(x, "medium", onnx=onnx_mode), "crepe-large": lambda: self.get_f0_crepe(x, "large", onnx=onnx_mode), "crepe-full": lambda: self.get_f0_crepe(x, "full", onnx=onnx_mode), "fcpe": lambda: self.get_f0_fcpe(x, p_len, int(hop_length), onnx=onnx_mode), "fcpe-legacy": lambda: self.get_f0_fcpe(x, p_len, int(hop_length), legacy=True, onnx=onnx_mode), "rmvpe": lambda: self.get_f0_rmvpe(x, onnx=onnx_mode), "rmvpe-legacy": lambda: self.get_f0_rmvpe(x, legacy=True, onnx=onnx_mode), "harvest": lambda: self.get_f0_pyworld(x, filter_radius, "harvest"), "yin": lambda: self.get_f0_yin(x, int(hop_length), p_len, mode="yin"), "pyin": lambda: self.get_f0_yin(x, int(hop_length), p_len, mode="pyin"), "swipe": lambda: self.get_f0_swipe(x)} f0 = self.get_f0_hybrid(f0_method, x, p_len, hop_length, filter_radius, onnx_mode) if "hybrid" in f0_method else f0_methods.get(f0_method, lambda: ValueError(translations["method_not_valid"]))() if f0_autotune: f0 = Autotune.autotune_f0(self, f0, f0_autotune_strength) if isinstance(f0, tuple): f0 = f0[0] f0 *= pow(2, pitch / 12) tf0 = self.sample_rate // self.window if inp_f0 is not None: replace_f0 = np.interp(list(range(np.round((inp_f0[:, 0].max() - inp_f0[:, 0].min()) * tf0 + 1).astype(np.int16))), inp_f0[:, 0] * 100, inp_f0[:, 1]) f0[self.x_pad * tf0 : self.x_pad * tf0 + len(replace_f0)] = replace_f0[:f0[self.x_pad * tf0 : self.x_pad * tf0 + len(replace_f0)].shape[0]] f0_mel = 1127 * np.log(1 + f0 / 700) f0_mel[f0_mel > 0] = (f0_mel[f0_mel > 0] - self.f0_mel_min) * 254 / (self.f0_mel_max - self.f0_mel_min) + 1 f0_mel[f0_mel <= 1] = 1 f0_mel[f0_mel > 255] = 255 return np.rint(f0_mel).astype(np.int32), f0.copy() def extract_features(self, model, feats, version): return torch.as_tensor(model.run([model.get_outputs()[0].name, model.get_outputs()[1].name], {"feats": feats.detach().cpu().numpy()})[0 if version == "v1" else 1], dtype=torch.float32, device=feats.device) def voice_conversion(self, model, net_g, sid, audio0, pitch, pitchf, index, big_npy, index_rate, version, protect): pitch_guidance = pitch != None and pitchf != None feats = (torch.from_numpy(audio0).half() if self.is_half else torch.from_numpy(audio0).float()) if feats.dim() == 2: feats = feats.mean(-1) assert feats.dim() == 1, feats.dim() feats = feats.view(1, -1) with torch.no_grad(): if self.embed_suffix == ".pt": padding_mask = torch.BoolTensor(feats.shape).to(self.device).fill_(False) logits = model.extract_features(**{"source": feats.to(self.device), "padding_mask": padding_mask, "output_layer": 9 if version == "v1" else 12}) feats = model.final_proj(logits[0]) if version == "v1" else logits[0] elif self.embed_suffix == ".onnx": feats = self.extract_features(model, feats.to(self.device), version).to(self.device) elif self.embed_suffix == ".safetensors": logits = model(feats.to(self.device))["last_hidden_state"] feats = (model.final_proj(logits[0]).unsqueeze(0) if version == "v1" else logits) else: raise ValueError(translations["option_not_valid"]) if protect < 0.5 and pitch_guidance: feats0 = feats.clone() if (not isinstance(index, type(None)) and not isinstance(big_npy, type(None)) and index_rate != 0): npy = feats[0].cpu().numpy() if self.is_half: npy = npy.astype(np.float32) score, ix = index.search(npy, k=8) weight = np.square(1 / score) npy = np.sum(big_npy[ix] * np.expand_dims(weight / weight.sum(axis=1, keepdims=True), axis=2), axis=1) if self.is_half: npy = npy.astype(np.float16) feats = (torch.from_numpy(npy).unsqueeze(0).to(self.device) * index_rate + (1 - index_rate) * feats) feats = F.interpolate(feats.permute(0, 2, 1), scale_factor=2).permute(0, 2, 1) if protect < 0.5 and pitch_guidance: feats0 = F.interpolate(feats0.permute(0, 2, 1), scale_factor=2).permute(0, 2, 1) p_len = audio0.shape[0] // self.window if feats.shape[1] < p_len: p_len = feats.shape[1] if pitch_guidance: pitch = pitch[:, :p_len] pitchf = pitchf[:, :p_len] if protect < 0.5 and pitch_guidance: pitchff = pitchf.clone() pitchff[pitchf > 0] = 1 pitchff[pitchf < 1] = protect pitchff = pitchff.unsqueeze(-1) feats = (feats * pitchff + feats0 * (1 - pitchff)).to(feats0.dtype) p_len = torch.tensor([p_len], device=self.device).long() audio1 = ((net_g.infer(feats.half() if self.is_half else feats.float(), p_len, pitch if pitch_guidance else None, (pitchf.half() if self.is_half else pitchf.float()) if pitch_guidance else None, sid)[0][0, 0]).data.cpu().float().numpy()) if self.suffix == ".pth" else (net_g.run([net_g.get_outputs()[0].name], ({net_g.get_inputs()[0].name: feats.cpu().numpy().astype(np.float32), net_g.get_inputs()[1].name: p_len.cpu().numpy(), net_g.get_inputs()[2].name: np.array([sid.cpu().item()], dtype=np.int64), net_g.get_inputs()[3].name: np.random.randn(1, 192, p_len).astype(np.float32), net_g.get_inputs()[4].name: pitch.cpu().numpy().astype(np.int64), net_g.get_inputs()[5].name: pitchf.cpu().numpy().astype(np.float32)} if pitch_guidance else {net_g.get_inputs()[0].name: feats.cpu().numpy().astype(np.float32), net_g.get_inputs()[1].name: p_len.cpu().numpy(), net_g.get_inputs()[2].name: np.array([sid.cpu().item()], dtype=np.int64), net_g.get_inputs()[3].name: np.random.randn(1, 192, p_len).astype(np.float32)}))[0][0, 0]) if self.embed_suffix == ".pt": del padding_mask del feats, p_len, net_g clear_gpu_cache() return audio1 def pipeline(self, model, net_g, sid, audio, pitch, f0_method, file_index, index_rate, pitch_guidance, filter_radius, volume_envelope, version, protect, hop_length, f0_autotune, f0_autotune_strength, suffix, embed_suffix, f0_file=None, f0_onnx=False, pbar=None): self.suffix = suffix self.embed_suffix = embed_suffix if file_index != "" and os.path.exists(file_index) and index_rate != 0: try: index = faiss.read_index(file_index) big_npy = index.reconstruct_n(0, index.ntotal) except Exception as e: logger.error(translations["read_faiss_index_error"].format(e=e)) index = big_npy = None else: index = big_npy = None pbar.update(1) opt_ts, audio_opt = [], [] audio = signal.filtfilt(bh, ah, audio) audio_pad = np.pad(audio, (self.window // 2, self.window // 2), mode="reflect") if audio_pad.shape[0] > self.t_max: audio_sum = np.zeros_like(audio) for i in range(self.window): audio_sum += audio_pad[i : i - self.window] for t in range(self.t_center, audio.shape[0], self.t_center): opt_ts.append(t - self.t_query + np.where(np.abs(audio_sum[t - self.t_query : t + self.t_query]) == np.abs(audio_sum[t - self.t_query : t + self.t_query]).min())[0][0]) s = 0 t, inp_f0 = None, None audio_pad = np.pad(audio, (self.t_pad, self.t_pad), mode="reflect") sid = torch.tensor(sid, device=self.device).unsqueeze(0).long() p_len = audio_pad.shape[0] // self.window if hasattr(f0_file, "name"): try: with open(f0_file.name, "r") as f: raw_lines = f.read() if len(raw_lines) > 0: inp_f0 = [] for line in raw_lines.strip("\n").split("\n"): inp_f0.append([float(i) for i in line.split(",")]) inp_f0 = np.array(inp_f0, dtype=np.float32) except: logger.error(translations["error_readfile"]) inp_f0 = None pbar.update(1) if pitch_guidance: pitch, pitchf = self.get_f0(audio_pad, p_len, pitch, f0_method, filter_radius, hop_length, f0_autotune, f0_autotune_strength, inp_f0, onnx_mode=f0_onnx) pitch, pitchf = pitch[:p_len], pitchf[:p_len] if self.device == "mps": pitchf = pitchf.astype(np.float32) pitch, pitchf = torch.tensor(pitch, device=self.device).unsqueeze(0).long(), torch.tensor(pitchf, device=self.device).unsqueeze(0).float() pbar.update(1) for t in opt_ts: t = t // self.window * self.window audio_opt.append(self.voice_conversion(model, net_g, sid, audio_pad[s : t + self.t_pad2 + self.window], pitch[:, s // self.window : (t + self.t_pad2) // self.window] if pitch_guidance else None, pitchf[:, s // self.window : (t + self.t_pad2) // self.window] if pitch_guidance else None, index, big_npy, index_rate, version, protect)[self.t_pad_tgt : -self.t_pad_tgt]) s = t audio_opt.append(self.voice_conversion(model, net_g, sid, audio_pad[t:], (pitch[:, t // self.window :] if t is not None else pitch) if pitch_guidance else None, (pitchf[:, t // self.window :] if t is not None else pitchf) if pitch_guidance else None, index, big_npy, index_rate, version, protect)[self.t_pad_tgt : -self.t_pad_tgt]) audio_opt = np.concatenate(audio_opt) if volume_envelope != 1: audio_opt = change_rms(audio, self.sample_rate, audio_opt, self.sample_rate, volume_envelope) audio_max = np.abs(audio_opt).max() / 0.99 if audio_max > 1: audio_opt /= audio_max if pitch_guidance: del pitch, pitchf del sid clear_gpu_cache() pbar.update(1) return audio_opt class VoiceConverter: def __init__(self, model_path, sid = 0): self.config = config self.device = config.device self.hubert_model = None self.tgt_sr = None self.net_g = None self.vc = None self.cpt = None self.version = None self.n_spk = None self.use_f0 = None self.loaded_model = None self.vocoder = "Default" self.checkpointing = False self.sample_rate = 16000 self.sid = sid self.get_vc(model_path, sid) def convert_audio(self, audio_input_path, audio_output_path, index_path, embedder_model, pitch, f0_method, index_rate, volume_envelope, protect, hop_length, f0_autotune, f0_autotune_strength, filter_radius, clean_audio, clean_strength, export_format, resample_sr = 0, checkpointing = False, f0_file = None, f0_onnx = False, embedders_mode = "fairseq", formant_shifting = False, formant_qfrency = 0.8, formant_timbre = 0.8, split_audio = False): try: with tqdm(total=10, desc=translations["convert_audio"], ncols=100, unit="a") as pbar: audio = load_audio(logger, audio_input_path, self.sample_rate, formant_shifting=formant_shifting, formant_qfrency=formant_qfrency, formant_timbre=formant_timbre) self.checkpointing = checkpointing audio_max = np.abs(audio).max() / 0.95 if audio_max > 1: audio /= audio_max pbar.update(1) if not self.hubert_model: models, _, embed_suffix = load_embedders_model(embedder_model, embedders_mode, providers=get_providers()) self.hubert_model = (models.to(self.device).half() if self.config.is_half else models.to(self.device).float()).eval() if embed_suffix in [".pt", ".safetensors"] else models self.embed_suffix = embed_suffix pbar.update(1) if self.tgt_sr != resample_sr >= self.sample_rate: self.tgt_sr = resample_sr target_sr = min([8000, 11025, 12000, 16000, 22050, 24000, 32000, 44100, 48000, 96000], key=lambda x: abs(x - self.tgt_sr)) if split_audio: chunks = cut(audio, self.sample_rate, db_thresh=-60, min_interval=500) pbar.total = len(chunks) * 4 + 6 logger.info(f"{translations['split_total']}: {len(chunks)}") else: chunks = [(audio, 0, 0)] converted_chunks = [] pbar.update(1) for waveform, start, end in chunks: converted_chunks.append((start, end, self.vc.pipeline(model=self.hubert_model, net_g=self.net_g, sid=self.sid, audio=waveform, pitch=pitch, f0_method=f0_method, file_index=(index_path.strip().strip('"').strip("\n").strip('"').strip().replace("trained", "added")), index_rate=index_rate, pitch_guidance=self.use_f0, filter_radius=filter_radius, volume_envelope=volume_envelope, version=self.version, protect=protect, hop_length=hop_length, f0_autotune=f0_autotune, f0_autotune_strength=f0_autotune_strength, suffix=self.suffix, embed_suffix=self.embed_suffix, f0_file=f0_file, f0_onnx=f0_onnx, pbar=pbar))) pbar.update(1) audio_output = restore(converted_chunks, total_len=len(audio), dtype=converted_chunks[0][2].dtype) if split_audio else converted_chunks[0][2] if target_sr >= self.sample_rate and self.tgt_sr != target_sr: audio_output = librosa.resample(audio_output, orig_sr=self.tgt_sr, target_sr=target_sr, res_type="soxr_vhq") pbar.update(1) if clean_audio: from main.tools.noisereduce import reduce_noise audio_output = reduce_noise(y=audio_output, sr=target_sr, prop_decrease=clean_strength, device=self.device) sf.write(audio_output_path, audio_output, target_sr, format=export_format) pbar.update(1) except Exception as e: logger.error(translations["error_convert"].format(e=e)) import traceback logger.debug(traceback.format_exc()) def get_vc(self, weight_root, sid): if sid == "" or sid == []: self.cleanup() clear_gpu_cache() if not self.loaded_model or self.loaded_model != weight_root: self.loaded_model = weight_root self.load_model() if self.cpt is not None: self.setup() def cleanup(self): if self.hubert_model is not None: del self.net_g, self.n_spk, self.vc, self.hubert_model, self.tgt_sr self.hubert_model = self.net_g = self.n_spk = self.vc = self.tgt_sr = None clear_gpu_cache() del self.net_g, self.cpt clear_gpu_cache() self.cpt = None def load_model(self): if os.path.isfile(self.loaded_model): if self.loaded_model.endswith(".pth"): self.cpt = torch.load(self.loaded_model, map_location="cpu") else: sess_options = onnxruntime.SessionOptions() sess_options.log_severity_level = 3 self.cpt = onnxruntime.InferenceSession(self.loaded_model, sess_options=sess_options, providers=get_providers()) else: self.cpt = None def setup(self): if self.cpt is not None: if self.loaded_model.endswith(".pth"): self.tgt_sr = self.cpt["config"][-1] self.cpt["config"][-3] = self.cpt["weight"]["emb_g.weight"].shape[0] self.use_f0 = self.cpt.get("f0", 1) self.version = self.cpt.get("version", "v1") self.vocoder = self.cpt.get("vocoder", "Default") if self.vocoder != "Default": self.config.is_half = False self.net_g = Synthesizer(*self.cpt["config"], use_f0=self.use_f0, text_enc_hidden_dim=768 if self.version == "v2" else 256, vocoder=self.vocoder, checkpointing=self.checkpointing) del self.net_g.enc_q self.net_g.load_state_dict(self.cpt["weight"], strict=False) self.net_g.eval().to(self.device) self.net_g = (self.net_g.half() if self.config.is_half else self.net_g.float()) self.n_spk = self.cpt["config"][-3] self.suffix = ".pth" else: import json import onnx metadata_dict = None for prop in onnx.load(self.loaded_model).metadata_props: if prop.key == "model_info": metadata_dict = json.loads(prop.value) break self.net_g = self.cpt self.tgt_sr = metadata_dict.get("sr", 32000) self.use_f0 = metadata_dict.get("f0", 1) self.version = metadata_dict.get("version", "v1") self.suffix = ".onnx" self.vc = VC(self.tgt_sr, self.config) if __name__ == "__main__": main()