音频提取特征_wav2vec2.0特征提取

目录

音频提取特征

音频切割

依赖项：

pip install librosa

pip install transformers

音频提取特征

import librosa
import numpy as np
import torch
from transformers import Wav2Vec2Processor
 
processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-base-960h")
 
 
 
wav_path = r'E:\迅雷下载\data\finedance\music_wav\001.wav'
speech_array, sampling_rate = librosa.load(wav_path, sr=16000)
input_values = np.squeeze(processor(speech_array, sampling_rate=16000).input_values)
 
print(input_values.shape)

音频切割

import librosa as lr
import numpy as np
import soundfile as sf
from tqdm import tqdm
 
 
def slice_audio(audio_file, stride, length, out_dir):
    # stride, length in seconds
    audio, sr = lr.load(audio_file, sr=None)
    file_name = os.path.splitext(os.path.basename(audio_file))[0]
    start_idx = 0
    idx = 0
    window = int(length * sr)
    stride_step = int(stride * sr)
    while start_idx <= len(audio) - window:
        audio_slice = audio[start_idx : start_idx + window]
        sf.write(f"{out_dir}/{file_name}_slice{idx}.wav", audio_slice, sr)
        start_idx += stride_step
        idx += 1
    return idx
 
 
def slice_audio_folder(wav_dir, stride=0.5, length=5):
    wavs = sorted(glob.glob(f"{wav_dir}/*.wav"))
    wav_out = wav_dir + "_sliced"
    os.makedirs(wav_out, exist_ok=True)
    for wav in tqdm(wavs):
        audio_slices = slice_audio(wav, stride, length, wav_out)
 
if __name__ == '__main__':
    slice_audio_folder(r"E:\Projects\FineDance-main\data\code")