音频去除前后静音以及音量归一化

实现两个功能：
1、基于音量的归一化
2、前后静音的切除

import os
import librosa
import pyloudnorm as pyln
import numpy as np
import soundfile as sf
import glob
import tqdm

sample_rate = 48000
trim_db = 30
wav_db = -20.0
st_save_len = 150 # ms
end_save_len = 200 # ms

meter = pyln.Meter(sample_rate) 


def process_wav(filename, file_o):
    wav, sr = librosa.load(filename, sr=sample_rate)
    loudness = meter.integrated_loudness(wav)
    wav = pyln.normalize.loudness(wav, loudness, wav_db)
    if np.abs(wav).max() > 1.0:
        wav = wav / np.abs(wav).max()

    wav = wav.astype(np.float32)
    cut_wav, index = librosa.effects.trim(
            wav,
            top_db=trim_db,
            frame_length=512,
            hop_length=128,
        )

    # index
    st_start, st_end = index[0], index[1]  # 这个可以用来进行手动调整静音长度,上面 cut_wav = wav[st_start, st_end]

    # new_s = max(st_start - int(st_save_len * sample_rate / 1000), 0)
    # new_d = min(st_end + int(end_save_len * sample_rate / 1000), len(wav))

    sf.write(file_o, cut_wav, sample_rate)