Python 转换wav->;mp3->;wav产生静态噪声
我正在尝试将我的麦克风音频转换成mp3文件,然后将它们保存在磁盘上,以便录制和保存音频曲目 但是要用Python 转换wav->;mp3->;wav产生静态噪声,python,python-3.x,audio,pyaudio,pydub,Python,Python 3.x,Audio,Pyaudio,Pydub,我正在尝试将我的麦克风音频转换成mp3文件,然后将它们保存在磁盘上,以便录制和保存音频曲目 但是要用pyaudio播放它,我需要将它转换成wav格式。 它最初是在wav中录制的,我正在尝试做wav->mp3->wav 我将代码的最小调试版本放在一起,如下所示: import pyaudio from array import array from struct import pack from sys import byteorder from io import BytesIO from py
pyaudio
播放它,我需要将它转换成wav
格式。它最初是在
wav
中录制的,我正在尝试做wav->mp3->wav
我将代码的最小调试版本放在一起,如下所示:
import pyaudio
from array import array
from struct import pack
from sys import byteorder
from io import BytesIO
from pydub import AudioSegment
p = pyaudio.PyAudio()
stream_mic = p.open(rate=11000,
format=pyaudio.paInt16,
channels=1,
input=True,
frames_per_buffer=500)
stream_out = p.open(rate=11000,
format=pyaudio.paInt16,
channels=1,
output=True,
frames_per_buffer=500)
def is_odd(a):
return bool(a - ((a>>1)<<1))
def wav_obj(raw_data):
wavHandle = AudioSegment(data=raw_data, sample_width=2, frame_rate=11000, channels=1)
return wavHandle
def wavToMp3(audioFrame):
mp3 = BytesIO()
file_handle = audioFrame.export(mp3, format="mp3")
mp3.seek(0)
data = mp3.read()
## == Data needs to be multiple of (sample_width * channels)
## Easiest way is to strip of a trailing data,
while is_odd(len(data)):
data = data[:-1]
return AudioSegment(data=data, sample_width=2, frame_rate=11000, channels=1)
def mp3ToWav(audioFrame):
#remasteredAudioFrame = audioFrame.set_frame_rate(11000)
wav = BytesIO()
file_handle = audioFrame.export(wav, format="wav")
wav.seek(0)
return AudioSegment(data=wav.read(), sample_width=2, frame_rate=11000, channels=1)
while 1:
snd_data = array('h', stream_mic.read(500))
if byteorder == 'big':
snd_data.byteswap()
frame = array('h')
frame.extend(snd_data)
wav = wav_obj(frame)
## == convert from .wav -> .mp3 -> .wav
## just to see the loss of audio.
mp3 = wavToMp3(wav)
wav = mp3ToWav(mp3)
stream_out.write(wav.raw_data)
stream_out.stop_stream()
stream_mic.stop_stream()
stream_out.close()
stream_mic.close()
p.terminate()
wav = wav_obj(frame)
sound = sound + wav
sound.export("test.mp3",
format="mp3",
bitrate="11k",
tags={"album": "test", "artist": "Not Ariana Grande"})
使播放声音“优美”。没有数据丢失,质量与我定义的一样
看在我的份上,我搞不清楚我在哪里把数据压缩搞混了。注意:如果可能的话,我希望在内存中执行此操作,因为我稍后将处理数据,以便尝试创建效果等 进展 正如@Anthon指出的,我应该一个接一个地隔离转换,看看哪一个会失败。我通过将
wav->mp3
保存到磁盘来实现这一点。started = time()
sound = AudioSegment(data=b'', sample_width=2, frame_rate=11000, channels=1)
while 1:
snd_data = array('h', stream_mic.read(500))
if byteorder == 'big':
snd_data.byteswap()
frame = array('h')
frame.extend(snd_data)
wav = wav_obj(frame)
## == convert from .wav -> .mp3 -> .wav
## just to see the loss of audio.
mp3 = wavToMp3(wav)
sound = sound + mp3
#wav = mp3ToWav(mp3)
#stream_out.write(mp3.raw_data)
if time() - started > 1.5:
break
print(sound.raw_data)
with open('test.mp3', 'wb') as fh:
fh.write(sound.raw_data)
然后我大胆地使用test.mp3
,看看波浪形成的样子。果然,看起来是
.mp3
转换不可靠
在肉眼看来,音频帧似乎被单独拖出和扭曲。声音总长度约为1.5秒,但从波形来看,压缩机添加的每一帧中都有暂停和延迟:
因此,在使用该值的所有实例中,我将值frames\u per\u buffer=500
增加到2000
:
stream_mic = p.open(frames_per_buffer=5000, ...)
stream_out = p.open(frames_per_buffer=5000, ...)
stream_mic.read(5000)
任何高于我的1.5秒的值都应该是足够好的值。
果然,波形看起来完全不同:
似乎有效的方法如下:
import pyaudio
from array import array
from struct import pack
from sys import byteorder
from io import BytesIO
from pydub import AudioSegment
p = pyaudio.PyAudio()
stream_mic = p.open(rate=11000,
format=pyaudio.paInt16,
channels=1,
input=True,
frames_per_buffer=500)
stream_out = p.open(rate=11000,
format=pyaudio.paInt16,
channels=1,
output=True,
frames_per_buffer=500)
def is_odd(a):
return bool(a - ((a>>1)<<1))
def wav_obj(raw_data):
wavHandle = AudioSegment(data=raw_data, sample_width=2, frame_rate=11000, channels=1)
return wavHandle
def wavToMp3(audioFrame):
mp3 = BytesIO()
file_handle = audioFrame.export(mp3, format="mp3")
mp3.seek(0)
data = mp3.read()
## == Data needs to be multiple of (sample_width * channels)
## Easiest way is to strip of a trailing data,
while is_odd(len(data)):
data = data[:-1]
return AudioSegment(data=data, sample_width=2, frame_rate=11000, channels=1)
def mp3ToWav(audioFrame):
#remasteredAudioFrame = audioFrame.set_frame_rate(11000)
wav = BytesIO()
file_handle = audioFrame.export(wav, format="wav")
wav.seek(0)
return AudioSegment(data=wav.read(), sample_width=2, frame_rate=11000, channels=1)
while 1:
snd_data = array('h', stream_mic.read(500))
if byteorder == 'big':
snd_data.byteswap()
frame = array('h')
frame.extend(snd_data)
wav = wav_obj(frame)
## == convert from .wav -> .mp3 -> .wav
## just to see the loss of audio.
mp3 = wavToMp3(wav)
wav = mp3ToWav(mp3)
stream_out.write(wav.raw_data)
stream_out.stop_stream()
stream_mic.stop_stream()
stream_out.close()
stream_mic.close()
p.terminate()
wav = wav_obj(frame)
sound = sound + wav
sound.export("test.mp3",
format="mp3",
bitrate="11k",
tags={"album": "test", "artist": "Not Ariana Grande"})
因此,wav-frame->mp3-frame->组合多个mp3-frame
无法工作。
但是执行wav帧->组合多个wav帧->导出到mp3
工作
这就是我再次陷入困境的地方。
显然,mp3
转换是不可靠的,它会在每个片段的波形中增加失真音调。
我希望在座的各位中有人曾经使用过pydub
,尤其是mp3转换,因为我完全迷路了
以下是收集的mp3:s:
我不知道.export
使用什么mp3编码器和/或是否可以设置其质量。我会先尝试进行外部转换(例如,lame
save wav>lame>read mp3),以确定是编码(很可能)而不是解码,然后看看如何提高罪犯的质量。@Anthon Solid tip!我将尝试使用外部编码器/解码器pyaudio
按顺序使用libav
或ffmpeg
。我试过这两种方法,结果都是一样的。我不认为你可以将MP3转换成块,然后再重新组合——大多数(可能全部?)MP3编码器会在转换后的音频的开头或结尾添加一些静音。如果您想压缩数据以节省带宽,请尝试通过lame管道传输数据,或仅抓住波形数据