读取 wav 格式声音文件

你的名字 2023-02-15 01:17 50阅读 0赞

#  读取 wav 格式声音文件 #

[http://bigsec.net/b52/scipydoc/wave\_pyaudio.html][http_bigsec.net_b52_scipydoc_wave_pyaudio.html]

Python 支持 wav 文件的读写，实时的声音输入输出需要安装 `pyAudio`，`pyMedia` 进行 MP3 的解码和播放。

wav 是 Microsoft 开发的一种声音文件格式，通常被用来保存未压缩的声音数据 （Pulse Code Modulation，PCM，脉冲编码调制)。wav 有三个重要的参数：声道数、采样频率和量化位数。

声道数：单声道 (mono) 或者是双声道 (stereo)。  
采样频率：每秒钟声音信号的采集次数。常用的有 8kHz、16kHz、32kHz、48kHz、11.025kHz、22.05kHz、44.1kHz 等。  
量化位数：用多少 bit 表达一次采样所采集的数据，通常有 8bit、16bit、24bit 和 32bit 等。CD 中所储存的声音信号是双声道、44.1kHz、16bit。

如果你需要自己录制和编辑声音文件，推荐使用 Audacity。它是一款开源的、跨平台、多声道的录音编辑软件。在工作中使用 Audacity 进行声音信号的录制，然后再输出成 wav 文件供 Python 程序处理。

## 1. `C:\Windows\media` ##

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2NoZW5neXExMTY_size_16_color_FFFFFF_t_70_pic_center]

(base) yongqiang@yongqiang:~$ cd /mnt/f/yongqiang_work/
    (base) yongqiang@yongqiang:/mnt/f/yongqiang_work$ ll
    total 260
    drwxrwxrwx 1 yongqiang yongqiang   4096 Jun  4 00:47 ./
    drwxrwxrwx 1 yongqiang yongqiang   4096 Jun  3 22:11 ../
    -rwxrwxrwx 1 yongqiang yongqiang 191788 Sep 15  2018 Windows_Ding.wav*
    -rwxrwxrwx 1 yongqiang yongqiang  70060 Sep 15  2018 ding.wav*
    (base) yongqiang@yongqiang:/mnt/f/yongqiang_work$
    (base) yongqiang@yongqiang:/mnt/f/yongqiang_work$ pwd
    /mnt/f/yongqiang_work
    (base) yongqiang@yongqiang:/mnt/f/yongqiang_work$

## 2. 读 wav 格式声音文件 ##

#!/usr/bin/env python
    # -*- coding: utf-8 -*-
    # yongqiang cheng
    
    from __future__ import absolute_import
    from __future__ import division
    from __future__ import print_function
    
    import wave
    import numpy as np
    
    # WAV file
    audio_file = "/mnt/f/yongqiang_work/ding.wav"
    object = wave.open(audio_file, "rb")
    
    # (nchannels, sampwidth, framerate, nframes, comptype, compname)
    params = object.getparams()
    nchannels, sampwidth, framerate, nframes, comptype, compname = params[:6]
    print("nchannels = ", nchannels)
    print("sampwidth = ", sampwidth)
    print("framerate = ", framerate)
    print("nframes = ", nframes)
    print("comptype = ", comptype)
    print("compname = ", compname)
    
    # Returns number of audio channels (1 for mono, 2 for stereo).
    print("object.getnchannels() = ", object.getnchannels())
    
    # Returns sample width in bytes.
    print("object.getsampwidth() = ", object.getsampwidth())
    
    # Returns sampling frequency.
    print("object.getframerate() = ", object.getframerate())
    
    # Returns number of audio frames.
    print("object.getnframes() = ", object.getnframes())
    
    # Returns compression type ('NONE' is the only supported type).
    print("object.getcomptype() = ", object.getcomptype())
    
    # Human-readable version of getcomptype(). Usually 'not compressed' parallels 'NONE'.
    print("object.getcompname() = ", object.getcompname())
    
    # Reads and returns at most n frames of audio, as a bytes object.
    str_data = object.readframes(nframes)
    object.close()

/home/yongqiang/miniconda3/envs/tf_cpu_1.4.1/bin/python /home/yongqiang/pycharm_work/yongqiang.py
    nchannels =  2
    sampwidth =  2
    framerate =  44100
    nframes =  17504
    comptype =  NONE
    compname =  not compressed
    object.getnchannels() =  2
    object.getsampwidth() =  2
    object.getframerate() =  44100
    object.getnframes() =  17504
    object.getcomptype() =  NONE
    object.getcompname() =  not compressed
    
    Process finished with exit code 0

## 3. 读 wav 格式声音文件 ##

#!/usr/bin/env python
    # -*- coding: utf-8 -*-
    # yongqiang cheng
    
    from __future__ import absolute_import
    from __future__ import division
    from __future__ import print_function
    
    import wave
    import numpy as np
    import matplotlib.pyplot as plt
    
    # WAV file
    audio_file = "/mnt/f/yongqiang_work/ding.wav"
    object = wave.open(audio_file, "rb")
    
    # (nchannels, sampwidth, framerate, nframes, comptype, compname)
    params = object.getparams()
    nchannels, sampwidth, framerate, nframes, comptype, compname = params[:6]
    print("nchannels =", nchannels)
    print("sampwidth =", sampwidth)
    print("framerate =", framerate)
    print("nframes =", nframes)
    print("comptype =", comptype)
    print("compname =", compname)
    
    # Returns number of audio channels (1 for mono, 2 for stereo).
    print("object.getnchannels() =", object.getnchannels())
    
    # Returns sample width in bytes.
    print("object.getsampwidth() =", object.getsampwidth())
    
    # Returns sampling frequency.
    print("object.getframerate() =", object.getframerate())
    
    # Returns number of audio frames.
    print("object.getnframes() =", object.getnframes())
    
    # Returns compression type ('NONE' is the only supported type).
    print("object.getcomptype() =", object.getcomptype())
    
    # Human-readable version of getcomptype(). Usually 'not compressed' parallels 'NONE'.
    print("object.getcompname() =", object.getcompname())
    
    # Reads and returns at most n frames of audio, as a bytes object.
    str_data = object.readframes(nframes)
    # nframes = 17504,  channels = 2, sampwidth = 2
    # str_data (bytes: 70016) = nframes * channels * sampwidth = 17504 * 2 * 2 = 70016
    object.close()
    
    wave_data = np.fromstring(str_data, dtype=np.short)
    wave_data.shape = -1, 2
    wave_data = wave_data.T
    time = np.arange(0, nframes) * (1.0 / framerate)
    
    plt.subplot(211)
    plt.plot(time, wave_data[0])
    plt.xlabel("left channel - time (seconds)")
    plt.subplot(212)
    plt.plot(time, wave_data[1], c="g")
    plt.xlabel("right channel - time (seconds)")
    plt.show()

/home/yongqiang/miniconda3/envs/pt-1.4_py-3.6/bin/python /home/yongqiang/pycharm_work/yongqiang.py
    nchannels = 2
    sampwidth = 2
    framerate = 44100
    nframes = 17504
    comptype = NONE
    compname = not compressed
    object.getnchannels() = 2
    object.getsampwidth() = 2
    object.getframerate() = 44100
    object.getnframes() = 17504
    object.getcomptype() = NONE
    object.getcompname() = not compressed
    
    Process finished with exit code 0

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2NoZW5neXExMTY_size_16_color_FFFFFF_t_70_pic_center 1]

Python 调用 `wave.open` 打开 wav 文件，注意需要使用 `"rb"` (二进制模式) 打开文件：

audio_file = "/mnt/f/yongqiang_work/ding.wav"
    object = wave.open(audio_file, "rb")

`open` 返回一个 `Wave_read` 类的实例，通过调用它的方法读取 wav 文件的格式和数据：

`getparams`：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。

# (nchannels, sampwidth, framerate, nframes, comptype, compname)
    params = object.getparams()
    nchannels, sampwidth, framerate, nframes, comptype, compname = params[:6]

`nchannels, sampwidth, framerate, nframes, comptype, compname` 等方法可以单独返回 wav 文件的特定的信息。

`readframes`：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)，`readframes` 返回的是二进制数据 (bytes)，在 Python 中用字符串表示二进制数据。

# Reads and returns at most n frames of audio, as a bytes object.
    str_data = object.readframes(nframes)
    # nframes = 17504,  channels = 2, sampwidth = 2
    # str_data (bytes: 70016) = nframes * channels * sampwidth = 17504 * 2 * 2 = 70016

接下来需要根据声道数和量化单位，将读取的二进制数据转换为一个可以计算的数组：

wave_data = np.fromstring(str_data, dtype=np.short)

通过 `fromstring` 函数将字符串转换为数组，通过其参数 `dtype` 指定转换后的数据格式，由于我们的声音格式是以两个字节表示一个取样值，因此采用 `short` 数据类型转换。现在得到的 `wave_data` 是一个一维的 `short` 类型的数组，但是因为我们的声音文件是双声道的，因此它由左右两个声道的取样交替构成：`LRLRLRLR....LR` (L 表示左声道的取样值，R 表示右声道取样值)。修改`wave_data` 的 `sharp` 之后：

wave_data.shape = -1, 2

将其转置得到：

wave_data = wave_data.T

最后通过取样点数和取样频率计算出每个取样的时间：

time = np.arange(0, nframes) * (1.0 / framerate)

## 4. sample width in bytes ##

#!/usr/bin/env python
    # -*- coding: utf-8 -*-
    # yongqiang cheng
    
    from __future__ import absolute_import
    from __future__ import division
    from __future__ import print_function
    
    import wave
    import numpy as np
    import matplotlib.pyplot as plt
    
    # WAV file
    audio_file = "/mnt/f/yongqiang_work/ding.wav"
    object = wave.open(audio_file, "rb")
    
    # (nchannels, sampwidth, framerate, nframes, comptype, compname)
    params = object.getparams()
    nchannels, sampwidth, framerate, nframes, comptype, compname = params[:6]
    print("nchannels =", nchannels)
    print("sampwidth =", sampwidth)
    print("framerate =", framerate)
    print("nframes =", nframes)
    print("comptype =", comptype)
    print("compname =", compname)
    
    # Returns number of audio channels (1 for mono, 2 for stereo).
    print("object.getnchannels() =", object.getnchannels())
    
    # Returns sample width in bytes.
    print("object.getsampwidth() =", object.getsampwidth())
    
    # Returns sampling frequency.
    print("object.getframerate() =", object.getframerate())
    
    # Returns number of audio frames.
    print("object.getnframes() =", object.getnframes())
    
    # Returns compression type ('NONE' is the only supported type).
    print("object.getcomptype() =", object.getcomptype())
    
    # Human-readable version of getcomptype(). Usually 'not compressed' parallels 'NONE'.
    print("object.getcompname() =", object.getcompname())
    
    # Reads and returns at most n frames of audio, as a bytes object.
    str_data = object.readframes(nframes)
    # nframes = 17504,  channels = 2, sampwidth = 2
    # str_data (bytes: 70016) = nframes * channels * sampwidth = 17504 * 2 * 2 = 70016
    num_bytes = len(str_data) # num_bytes = 70016
    print("num_bytes =", num_bytes, "bytes")
    object.close()
    
    wave_data = np.fromstring(str_data, dtype=np.short)
    wave_data.shape = -1, 2
    wave_data = wave_data.T
    time = np.arange(0, nframes) * (1.0 / framerate)
    
    plt.subplot(211)
    plt.plot(time, wave_data[0])
    plt.xlabel("left channel - time (seconds)")
    plt.subplot(212)
    plt.plot(time, wave_data[1], c="g")
    plt.xlabel("right channel - time (seconds)")
    plt.show()

/home/yongqiang/miniconda3/envs/pt-1.4_py-3.6/bin/python /home/yongqiang/pytorch_work/end2end-asr-pytorch-example/yongqiang.py
    nchannels = 2
    sampwidth = 2
    framerate = 44100
    nframes = 17504
    comptype = NONE
    compname = not compressed
    object.getnchannels() = 2
    object.getsampwidth() = 2
    object.getframerate() = 44100
    object.getnframes() = 17504
    object.getcomptype() = NONE
    object.getcompname() = not compressed
    num_bytes = 70016 bytes
    
    Process finished with exit code 0

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2NoZW5neXExMTY_size_16_color_FFFFFF_t_70_pic_center 2]

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2NoZW5neXExMTY_size_16_color_FFFFFF_t_70_pic_center 3]

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2NoZW5neXExMTY_size_16_color_FFFFFF_t_70_pic_center 4]

## References ##

[http://bigsec.net/b52/scipydoc/wave\_pyaudio.html][http_bigsec.net_b52_scipydoc_wave_pyaudio.html]

[http_bigsec.net_b52_scipydoc_wave_pyaudio.html]: http://bigsec.net/b52/scipydoc/wave_pyaudio.html
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2NoZW5neXExMTY_size_16_color_FFFFFF_t_70_pic_center]: https://img-blog.csdnimg.cn/20200604003922298.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2NoZW5neXExMTY=,size_16,color_FFFFFF,t_70#pic_center
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2NoZW5neXExMTY_size_16_color_FFFFFF_t_70_pic_center 1]: https://img-blog.csdnimg.cn/2020060710390167.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2NoZW5neXExMTY=,size_16,color_FFFFFF,t_70#pic_center
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2NoZW5neXExMTY_size_16_color_FFFFFF_t_70_pic_center 2]: https://img-blog.csdnimg.cn/20200704155855256.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2NoZW5neXExMTY=,size_16,color_FFFFFF,t_70#pic_center
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2NoZW5neXExMTY_size_16_color_FFFFFF_t_70_pic_center 3]: https://img-blog.csdnimg.cn/20200704160158922.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2NoZW5neXExMTY=,size_16,color_FFFFFF,t_70#pic_center
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2NoZW5neXExMTY_size_16_color_FFFFFF_t_70_pic_center 4]: https://img-blog.csdnimg.cn/20200704160319693.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2NoZW5neXExMTY=,size_16,color_FFFFFF,t_70#pic_center