The audio processing related knowledge that I encountered in the learning process

PCM file

.pcm : PCM (Pulse-code modulation) -> Pulse-code modulation. PCM can convert the sound from analog signal to digital signal. The principle is: use a fixed frequency to sample the analog signal, and the sampled signal looks like a series of continuous pulses with different amplitudes on the waveform. The amplitude is quantized according to a certain precision , and these quantized values ​​are continuously output, transmitted, processed or recorded in the storage medium, which constitutes the generation process of digital audio.
PCM data is completely lossless, with excellent sound quality but bulky.
Compression method: lossless compression (ALAC, APE, FLAC)
lossy compression (MP3, AAC, OGG, WMA)

1. Understand audio

1. Get audio information

def wav_infos(wav_path):
    '''
    获取音频信息

    :param wav_path: 音频路径
    :return: [1, 2, 8000, 51158, 'NONE', 'not compressed']
    对应关系：声道，采样宽度，帧速率，帧数，唯一标识，无损
    '''
    with wave.open(wav_path, "rb") as f:
        f = wave.open(wav_path)

        return list(f.getparams())

2. Read the content of the audio file

def read_wav(wav_path):
    '''
    读取音频文件内容:只能读取单声道的音频文件, 这个比较耗时

    :param wav_path: 音频路径
    :return:  音频内容
    '''
    with wave.open(wav_path, "rb") as f:
        # 读取格式信息
        # 一次性返回所有的WAV文件的格式信息，它返回的是一个组元(tuple)：声道数, 量化位数（byte单位）, 采
        # 样频率, 采样点数, 压缩类型, 压缩类型的描述。wave模块只支持非压缩的数据，因此可以忽略最后两个信息
        params = f.getparams()
        nchannels, sampwidth, framerate, nframes = params[:4]

        # 读取声音数据，传递一个参数指定需要读取的长度（以取样点为单位）
        str_data = f.readframes(nframes)

    return str_data

3. Get the audio duration (seconds)

def get_wav_time(wav_path):
    '''
    获取音频文件是时长

    :param wav_path: 音频路径
    :return: 音频时长 (单位秒)
    '''
    with contextlib.closing(wave.open(wav_path, 'r')) as f:
        frames = f.getnframes()
    rate = f.getframerate()
    duration = frames / float(rate)
    return duration

4. Audio Slicing

# 音频切片，获取部分音频 时间的单位是毫秒
start_time = 13950
end_time = 15200
get_ms_part_wav(main_wav_path, start_time, end_time, part_wav_path)


# 音频切片，获取部分音频 时间的单位是秒
start_time = 35
end_time = 38
get_second_part_wav(main_wav_path, start_time, end_time, second_part_wav_path)

# 音频切片，获取部分音频 时间的单位是分钟和秒 样式：0:12
start_time = "0:35"
end_time = "0:38"
get_minute_part_wav(main_wav_path, start_time, end_time, minute_part_wav_path)

5. Mutual conversion between wav files and pcm files

# wav文件转为pcm文件
wav_to_pcm(wav_path, pcm_path)

# pcm文件转为wav文件
pcm_to_wav(pcm_path, wav_path2)

6. Draw the waveform corresponding to the audio
wav_waveform(wav_path)

def wav_waveform(wave_path):
    '''
    音频对应的波形图
    :param wave_path:  音频路径
    :return:
    '''
    file = wave.open(wave_path)
    # print('---------声音信息------------')
    # for item in enumerate(WAVE.getparams()):
    #     print(item)
    a = file.getparams().nframes  # 帧总数
    f = file.getparams().framerate  # 采样频率
    sample_time = 1 / f  # 采样点的时间间隔
    time = a / f  # 声音信号的长度
    sample_frequency, audio_sequence = wavfile.read(wave_path)
    # print(audio_sequence)  # 声音信号每一帧的“大小”
    x_seq = np.arange(0, time, sample_time)

    plt.plot(x_seq, audio_sequence, 'blue')
    plt.xlabel("time (s)")
    plt.show()

The code call comes from the waveTools toolkit (for reference)

# -*- coding:utf8 -*-
'''
auth： Young
公众号：Python疯子 （Hold2Crazy）
'''
import wave
import contextlib
import numpy as np
import matplotlib.pyplot as plt

from scipy.io import wavfile
from pydub import AudioSegment


def wav_infos(wav_path):
    '''
    获取音频信息

    :param wav_path: 音频路径
    :return: [1, 2, 8000, 51158, 'NONE', 'not compressed']
    对应关系：声道，采样宽度，帧速率，帧数，唯一标识，无损
    '''
    with wave.open(wav_path, "rb") as f:
        f = wave.open(wav_path)

        return list(f.getparams())

def read_wav(wav_path):
    '''
    读取音频文件内容:只能读取单声道的音频文件, 这个比较耗时

    :param wav_path: 音频路径
    :return:  音频内容
    '''
    with wave.open(wav_path, "rb") as f:
        # 读取格式信息
        # 一次性返回所有的WAV文件的格式信息，它返回的是一个组元(tuple)：声道数, 量化位数（byte单位）, 采
        # 样频率, 采样点数, 压缩类型, 压缩类型的描述。wave模块只支持非压缩的数据，因此可以忽略最后两个信息
        params = f.getparams()
        nchannels, sampwidth, framerate, nframes = params[:4]

        # 读取声音数据，传递一个参数指定需要读取的长度（以取样点为单位）
        str_data = f.readframes(nframes)

    return str_data

def get_wav_time(wav_path):
    '''
    获取音频文件是时长

    :param wav_path: 音频路径
    :return: 音频时长 (单位秒)
    '''
    with contextlib.closing(wave.open(wav_path, 'r')) as f:
        frames = f.getnframes()
    rate = f.getframerate()
    duration = frames / float(rate)
    return duration


def get_ms_part_wav(main_wav_path, start_time, end_time, part_wav_path):
    '''
    音频切片，获取部分音频 单位是毫秒级别

    :param main_wav_path: 原音频文件路径
    :param start_time:  截取的开始时间
    :param end_time:  截取的结束时间
    :param part_wav_path:  截取后的音频路径
    :return:
    '''
    start_time = int(start_time)
    end_time = int(end_time)

    sound = AudioSegment.from_mp3(main_wav_path)
    word = sound[start_time:end_time]

    word.export(part_wav_path, format="wav")


def get_second_part_wav(main_wav_path, start_time, end_time, part_wav_path):
    '''
    音频切片，获取部分音频 单位是秒级别

    :param main_wav_path: 原音频文件路径
    :param start_time:  截取的开始时间
    :param end_time:  截取的结束时间
    :param part_wav_path:  截取后的音频路径
    :return:
    '''
    start_time = int(start_time) * 1000
    end_time = int(end_time) * 1000

    sound = AudioSegment.from_mp3(main_wav_path)
    word = sound[start_time:end_time]

    word.export(part_wav_path, format="wav")

def get_minute_part_wav(main_wav_path, start_time, end_time, part_wav_path):
    '''
    音频切片，获取部分音频 分钟:秒数  时间样式："12:35"

    :param main_wav_path: 原音频文件路径
    :param start_time:  截取的开始时间
    :param end_time:  截取的结束时间
    :param part_wav_path:  截取后的音频路径
    :return:
    '''

    start_time = (int(start_time.split(':')[0])*60+int(start_time.split(':')[1]))*1000
    end_time = (int(end_time.split(':')[0])*60+int(end_time.split(':')[1]))*1000

    sound = AudioSegment.from_mp3(main_wav_path)
    word = sound[start_time:end_time]

    word.export(part_wav_path, format="wav")


def wav_to_pcm(wav_path, pcm_path):
    '''
    wav文件转为pcm文件

    :param wav_path:wav文件路径
    :param pcm_path:要存储的pcm文件路径
    :return: 返回结果
    '''
    f = open(wav_path, "rb")
    f.seek(0)
    f.read(44)

    data = np.fromfile(f, dtype=np.int16)
    data.tofile(pcm_path)

def pcm_to_wav(pcm_path, wav_path):
    '''
    pcm文件转为wav文件

    :param pcm_path: pcm文件路径
    :param wav_path: wav文件路径
    :return:
    '''
    f = open(pcm_path,'rb')
    str_data  = f.read()
    wave_out=wave.open(wav_path,'wb')
    wave_out.setnchannels(1)
    wave_out.setsampwidth(2)
    wave_out.setframerate(8000)
    wave_out.writeframes(str_data)

# 音频对应的波形图
def wav_waveform(wave_path):
    '''
    音频对应的波形图
    :param wave_path:  音频路径
    :return:
    '''
    file = wave.open(wave_path)
    # print('---------声音信息------------')
    # for item in enumerate(WAVE.getparams()):
    #     print(item)
    a = file.getparams().nframes  # 帧总数
    f = file.getparams().framerate  # 采样频率
    sample_time = 1 / f  # 采样点的时间间隔
    time = a / f  # 声音信号的长度
    sample_frequency, audio_sequence = wavfile.read(wave_path)
    # print(audio_sequence)  # 声音信号每一帧的“大小”
    x_seq = np.arange(0, time, sample_time)

    plt.plot(x_seq, audio_sequence, 'blue')
    plt.xlabel("time (s)")
    plt.show()

2. Processing audio (intercepting audio)

method one:

from pydub import AudioSegment
import os
import librosa
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import math

# 显示音频时域波形
file_path = 'E:\shipsEar_AUDIOS-20220902T005604Z-001\shipsEar_AUDIOS\96__A__Draga_4.wav'
fname, ext = os.path.split(file_path)

input_music = AudioSegment.from_wav(file_path)

# 开始截取时间
start = 0 * 1000
end = 60 *1000
number = math.floor((end-start)/(5*1000))
for i in range(number):
     output_music = input_music[(start + i* 5*1000):(start + i*5*1000 + 5*1000)]
     output_music.export('E:\shipsEar_AUDIOS-20220902T005604Z-001\Data_frame\96__A__Draga_' + str(i+1) + '.wav', format="wav")

Method Two

Put all the audio in the same folder, but file naming can become an issue

import os
from pydub import AudioSegment
import numpy as np
########################开始批处理######################
file_path1 = r"E:\copy2"#输入路径
file_path2 = r"E:\dataset1"  #输出路径
for filepath,dirnames,filenames in os.walk(file_path1):
    for file in filenames: #遍历文件
        path1 = filepath+'\\'+file
        filename = file.split('.')[0] #不带 .wav的文件名
    ########################处理音频文件#######################
        audio = AudioSegment.from_file(path1, "wav")
        audio_time = len(audio)  # 获取待切割音频的时长，单位是毫秒
        cut_parameters = np.arange(0.3, audio_time / 1000, 0.3)  # np.arange()函数第一个参数为起点，第二个参数为终点，第三个参数为步长（10秒）
        start_time = int(0)  # 开始时间设为0
    ########################根据数组切割音频####################
        for t in cut_parameters:
            stop_time = int(t * 1000)  # pydub以毫秒为单位工作
            # print(stop_time)
            audio_chunk = audio[start_time:stop_time]  # 音频切割按开始时间到结束时间切割
            print("split at [{}:{}] ms".format(start_time, stop_time))
            audio_chunk.export(file_path2 + '\\' + filename + "-{}.wav".format(int(t*10)), format="wav")  #保存音频文件
            start_time = stop_time
            print('finish')