Python处理音频

从video中抽取audio

from moviepy.editor import VideoFileClip
from pydub import AudioSegment

video_path = '/opt/audio/audios/video1.mp4'
audio_path = '/opt/audio/audios/video1.wav'  # 提取的音频保存路径

# 加载视频文件
video = VideoFileClip(video_path)

# 提取音频
audio = video.audio

# 保存音频文件
audio.write_audiofile(audio_path)


# 读取音频文件
sound = AudioSegment.from_wav(audio_path)

# 将音频转换为单声道
sound = sound.set_channels(1)
sound = sound.set_frame_rate(16000)

# 保存音频文件（单声道）

sound.export(audio_path, format="wav")

截取audio

from pydub import AudioSegment

# 加载音频文件
audio = AudioSegment.from_file(
    "xxx/1.wav")

# 定义起始和结束时间（单位为毫秒）
start_time = 3000  
end_time = 28550   

# 截取音频
extracted = audio[start_time:end_time]

# 导出截取的音频
extracted.export(
    "xxx/3.wav", format="wav")

拼接audio

import os
from pydub import AudioSegment


base_path = f'{os.getcwd()}/audios/reduce_noise/video2/'
# 要拼接的音频的路径list
short_audio_files = []
for i in range(100, 105):
    path = base_path + str(i) + ".wav"
    wav = AudioSegment.from_file(path)
    duration_seconds = wav.duration_seconds
    short_audio_files.append(path)
# 声明一个空白音频
merged_audio = AudioSegment.empty()
# 遍历每个短音频文件并添加到合并后的音频中
for audio_file in short_audio_files:
    # 从文件加载短音频
    short_audio = AudioSegment.from_file(audio_file)
    # 将短音频追加到合并后的音频
    merged_audio = merged_audio.append(short_audio, crossfade=0)
    
# 保存合并后的音频为一个长音频文件
merged_audio.export(f"{base_path}merged_audio.wav", format="wav")

python绘制梅尔谱图像

import os
import matplotlib
 
matplotlib.use('Agg')  # No pictures displayed
import pylab
import librosa
import librosa.display
import numpy as np
 
sig, fs = librosa.load(os.path.join(os.path.dirname(os.path.realpath(__file__)), "video1.wav"))
# make pictures name
save_path = 'test.jpg'
 
# pylab.axis('off')  # no axis
# pylab.axes([0., 0., 1., 1.], frameon=False, xticks=[], yticks=[])  # Remove the white edge
S = librosa.feature.melspectrogram(y=sig, sr=fs)
librosa.display.specshow(librosa.power_to_db(S, ref=np.max),y_axis="mel", fmax=8000 * 1, x_axis="time")
pylab.savefig(save_path, bbox_inches=None, pad_inches=0)
pylab.close()

参考文章： 

02｜如何量化分析语音信号？-极客时间 

说话人分离：

主要使用whisper-diarization：GitHub - MahmoudAshraf97/whisper-diarization: Automatic Speech Recognition with Speaker Diarization based on OpenAI Whisper

使用场景：一整个wav ，分理出不同人对应的wav，具体效果如下：

 语音活性检测（VAD）

使用webrtc自带的VAD（贼好用）