rvc/train/preprocess/preprocess.py

import os
import sys
import time
from scipy import signal
from scipy.io import wavfile
import numpy as np
import concurrent.futures
from tqdm import tqdm
import json
from distutils.util import strtobool
import librosa
import multiprocessing
import noisereduce as nr
import soxr

now_directory = os.getcwd()
sys.path.append(now_directory)

from rvc.lib.utils import load_audio
from rvc.train.preprocess.slicer import Slicer

import logging

logging.getLogger("numba.core.byteflow").setLevel(logging.WARNING)
logging.getLogger("numba.core.ssa").setLevel(logging.WARNING)
logging.getLogger("numba.core.interpreter").setLevel(logging.WARNING)

OVERLAP = 0.3
PERCENTAGE = 3.0
MAX_AMPLITUDE = 0.9
ALPHA = 0.75
HIGH_PASS_CUTOFF = 48
SAMPLE_RATE_16K = 16000
RES_TYPE = "soxr_vhq"


class PreProcess:
    def __init__(self, sr: int, exp_dir: str):
        self.slicer = Slicer(
            sr=sr,
            threshold=-42,
            min_length=1500,
            min_interval=400,
            hop_size=15,
            max_sil_kept=500,
        )
        self.sr = sr
        self.b_high, self.a_high = signal.butter(
            N=5, Wn=HIGH_PASS_CUTOFF, btype="high", fs=self.sr
        )
        self.exp_dir = exp_dir
        self.device = "cpu"
        self.gt_wavs_dir = os.path.join(exp_dir, "sliced_audios")
        self.wavs16k_dir = os.path.join(exp_dir, "sliced_audios_16k")
        os.makedirs(self.gt_wavs_dir, exist_ok=True)
        os.makedirs(self.wavs16k_dir, exist_ok=True)

    def _normalize_audio(self, audio: np.ndarray):
        tmp_max = np.abs(audio).max()
        if tmp_max > 2.5:
            return None
        return (audio / tmp_max * (MAX_AMPLITUDE * ALPHA)) + (1 - ALPHA) * audio

    def process_audio_segment(
        self,
        normalized_audio: np.ndarray,
        sid: int,
        idx0: int,
        idx1: int,
        normalization_mode: str,
    ):
        if normalized_audio is None:
            print(f"{sid}-{idx0}-{idx1}-filtered")
            return
        if normalization_mode == "post":
            normalized_audio = self._normalize_audio(normalized_audio)
        wavfile.write(
            os.path.join(self.gt_wavs_dir, f"{sid}_{idx0}_{idx1}.wav"),
            self.sr,
            normalized_audio.astype(np.float32),
        )
        audio_16k = librosa.resample(
            normalized_audio,
            orig_sr=self.sr,
            target_sr=SAMPLE_RATE_16K,
            res_type=RES_TYPE,
        )
        wavfile.write(
            os.path.join(self.wavs16k_dir, f"{sid}_{idx0}_{idx1}.wav"),
            SAMPLE_RATE_16K,
            audio_16k.astype(np.float32),
        )

    def simple_cut(
        self,
        audio: np.ndarray,
        sid: int,
        idx0: int,
        chunk_len: float,
        overlap_len: float,
        normalization_mode: str,
    ):
        chunk_length = int(self.sr * chunk_len)
        overlap_length = int(self.sr * overlap_len)
        i = 0
        while i < len(audio):
            chunk = audio[i : i + chunk_length]
            if normalization_mode == "post":
                chunk = self._normalize_audio(chunk)
            if len(chunk) == chunk_length:
                # full SR for training
                wavfile.write(
                    os.path.join(
                        self.gt_wavs_dir,
                        f"{sid}_{idx0}_{i // (chunk_length - overlap_length)}.wav",
                    ),
                    self.sr,
                    chunk.astype(np.float32),
                )
                # 16KHz for feature extraction
                chunk_16k = librosa.resample(
                    chunk, orig_sr=self.sr, target_sr=SAMPLE_RATE_16K, res_type=RES_TYPE
                )
                wavfile.write(
                    os.path.join(
                        self.wavs16k_dir,
                        f"{sid}_{idx0}_{i // (chunk_length - overlap_length)}.wav",
                    ),
                    SAMPLE_RATE_16K,
                    chunk_16k.astype(np.float32),
                )
            i += chunk_length - overlap_length

    def process_audio(
        self,
        path: str,
        idx0: int,
        sid: int,
        cut_preprocess: str,
        process_effects: bool,
        noise_reduction: bool,
        reduction_strength: float,
        chunk_len: float,
        overlap_len: float,
        normalization_mode: str,
    ):
        audio_length = 0
        try:
            audio = load_audio(path, self.sr)
            audio_length = librosa.get_duration(y=audio, sr=self.sr)

            if process_effects:
                audio = signal.lfilter(self.b_high, self.a_high, audio)
            if normalization_mode == "pre":
                audio = self._normalize_audio(audio)
            if noise_reduction:
                audio = nr.reduce_noise(
                    y=audio, sr=self.sr, prop_decrease=reduction_strength
                )
            if cut_preprocess == "Skip":
                # no cutting
                self.process_audio_segment(
                    audio,
                    sid,
                    idx0,
                    0,
                    normalization_mode,
                )
            elif cut_preprocess == "Simple":
                # simple
                self.simple_cut(
                    audio,
                    sid,
                    idx0,
                    chunk_len,
                    overlap_len,
                    normalization_mode,
                )
            elif cut_preprocess == "Automatic":
                idx1 = 0
                # legacy
                for audio_segment in self.slicer.slice(audio):
                    i = 0
                    while True:
                        start = int(self.sr * (PERCENTAGE - OVERLAP) * i)
                        i += 1
                        if (
                            len(audio_segment[start:])
                            > (PERCENTAGE + OVERLAP) * self.sr
                        ):
                            tmp_audio = audio_segment[
                                start : start + int(PERCENTAGE * self.sr)
                            ]
                            self.process_audio_segment(
                                tmp_audio,
                                sid,
                                idx0,
                                idx1,
                                normalization_mode,
                            )
                            idx1 += 1
                        else:
                            tmp_audio = audio_segment[start:]
                            self.process_audio_segment(
                                tmp_audio,
                                sid,
                                idx0,
                                idx1,
                                normalization_mode,
                            )
                            idx1 += 1
                            break

        except Exception as error:
            print(f"Error processing audio: {error}")
        return audio_length


def format_duration(seconds):
    hours = int(seconds // 3600)
    minutes = int((seconds % 3600) // 60)
    seconds = int(seconds % 60)
    return f"{hours:02}:{minutes:02}:{seconds:02}"


def save_dataset_duration(file_path, dataset_duration):
    try:
        with open(file_path, "r") as f:
            data = json.load(f)
    except FileNotFoundError:
        data = {}

    formatted_duration = format_duration(dataset_duration)
    new_data = {
        "total_dataset_duration": formatted_duration,
        "total_seconds": dataset_duration,
    }
    data.update(new_data)

    with open(file_path, "w") as f:
        json.dump(data, f, indent=4)


def process_audio_wrapper(args):
    (
        pp,
        file,
        cut_preprocess,
        process_effects,
        noise_reduction,
        reduction_strength,
        chunk_len,
        overlap_len,
        normalization_mode,
    ) = args
    file_path, idx0, sid = file
    return pp.process_audio(
        file_path,
        idx0,
        sid,
        cut_preprocess,
        process_effects,
        noise_reduction,
        reduction_strength,
        chunk_len,
        overlap_len,
        normalization_mode,
    )


def preprocess_training_set(
    input_root: str,
    sr: int,
    num_processes: int,
    exp_dir: str,
    cut_preprocess: str,
    process_effects: bool,
    noise_reduction: bool,
    reduction_strength: float,
    chunk_len: float,
    overlap_len: float,
    normalization_mode: str,
):
    start_time = time.time()
    pp = PreProcess(sr, exp_dir)
    print(f"Starting preprocess with {num_processes} processes...")

    files = []
    idx = 0

    for root, _, filenames in os.walk(input_root):
        try:
            sid = 0 if root == input_root else int(os.path.basename(root))
            for f in filenames:
                if f.lower().endswith((".wav", ".mp3", ".flac", ".ogg")):
                    files.append((os.path.join(root, f), idx, sid))
                    idx += 1
        except ValueError:
            print(
                f'Speaker ID folder is expected to be integer, got "{os.path.basename(root)}" instead.'
            )

    # print(f"Number of files: {len(files)}")
    audio_length = []
    with tqdm(total=len(files)) as pbar:
        with concurrent.futures.ProcessPoolExecutor(
            max_workers=num_processes
        ) as executor:
            futures = [
                executor.submit(
                    process_audio_wrapper,
                    (
                        pp,
                        file,
                        cut_preprocess,
                        process_effects,
                        noise_reduction,
                        reduction_strength,
                        chunk_len,
                        overlap_len,
                        normalization_mode,
                    ),
                )
                for file in files
            ]
            for future in concurrent.futures.as_completed(futures):
                audio_length.append(future.result())
                pbar.update(1)

    audio_length = sum(audio_length)
    save_dataset_duration(
        os.path.join(exp_dir, "model_info.json"), dataset_duration=audio_length
    )
    elapsed_time = time.time() - start_time
    print(
        f"Preprocess completed in {elapsed_time:.2f} seconds on {format_duration(audio_length)} seconds of audio."
    )


if __name__ == "__main__":
    experiment_directory = str(sys.argv[1])
    input_root = str(sys.argv[2])
    sample_rate = int(sys.argv[3])
    num_processes = sys.argv[4]
    if num_processes.lower() == "none":
        num_processes = multiprocessing.cpu_count()
    else:
        num_processes = int(num_processes)
    cut_preprocess = str(sys.argv[5])
    process_effects = strtobool(sys.argv[6])
    noise_reduction = strtobool(sys.argv[7])
    reduction_strength = float(sys.argv[8])
    chunk_len = float(sys.argv[9])
    overlap_len = float(sys.argv[10])
    normalization_mode = str(sys.argv[11])
    preprocess_training_set(
        input_root,
        sample_rate,
        num_processes,
        experiment_directory,
        cut_preprocess,
        process_effects,
        noise_reduction,
        reduction_strength,
        chunk_len,
        overlap_len,
        normalization_mode,
    )