使用librosa的STFT理解

时间:2019-07-16 13:51:07

标签: python audio fft sampling librosa

我有一个8khz采样率的音频采样,大约14秒。 我正在使用librosa从该音频文件中提取一些功能。

y, sr = librosa.load(file_name)
stft = np.abs(librosa.stft(y, n_fft=n_fft))

# file_length = 14.650022675736961 #sec
# defaults 
# n_fft =2048
# hop_length = 512 # win_length/4 = n_fft/4 = 512 (win_length = n_fft default)

#windowsTime = n_fft * Ts # (1/sr)

stft.shape
# (1025, 631)

Specshow:

librosa.display.specshow(stft, x_axis='time', y_axis='log')

[![stft sr = 22050] [1]] [1]

现在, 我可以理解STFT的形状

631 time bins = are 4 * ( file_length / Ts * windowsTime) #overlapping
1025 frequency bins = Frames frequency gap sr/n_fft.
so there are 1025 frequencies in 0 to sr/2(Nyquest)

我不明白的是两种不同采样率的不同情节 具有相同的比例。 1-22050作为默认的librosa 2-8khz作为采样率文件

y2, sr = librosa.load(file_name, sr=None)

n_fft2 =743 # (same ratio to get same visuals for comparsion)
hop_length = 186 # (1/4 n_fft by default)

stft2 = np.abs(librosa.stft(y2, n_fft=n_fft2))

因此,stft的威信会有所不同

stft2.shape
# (372, 634)


[![stft sr = 743] [2]] [2]

1。但是为什么绝对频率不一样呢?其相同的信号只是不被过采样,因此每个采样都是唯一的。 我想念什么?是静态的y轴吗?

2。我无法理解时间仓值。我希望从该点到文件末尾的第一个为跳数长度,第二个为windowTime时,帧数为bin。但是单位很奇怪?

我希望能够在特定的时间(帧)中提取特定Fbin的幅度,或者能够对其中的一些求和以求出时间范围的磁化强度。

因此,如果我将stft [fBin的数量]取为1025 fBins的1行(stft [1025])并查看其内容,则stft [0]包含630点,对于每个频率而言,这正是630个时间点因此每帧1-1025的时间点都相同。

因此,如果我也采样了一个适合所有其他fbin的样本(相同时间点),即stft [0] 我将能够选择时间范围和fBin并获得特定的幅度:

times =  librosa.core.frames_to_time(stft2[0], sr=sr2, n_fft=n_fft2, hop_length=hop_length) 

fft_bin = 6
time_idx = 10

print('freq (Hz)', freqs[fft_bin])
print('time (s)', times[time_idx])
print('amplitude', stft[fft_bin, time_idx])

array([0.047375,0.047625,0.04825,0.04825,0.046875,0.04675,        0.05,0.051625,0.051,0.048,0.05225,0.050375,        0.04925,0.04725,0.051625,0.0465,0.05225,0.05,        0.053,0.053875,0.048,0.0485,0.047875,0.04775,        0.0485、0.049、0.051375、0.047125、0.051125、0.047125,        0.04725、0.05025、0.05425、0.05475、0.051375、0.060375,        0.050625、0.04875、0.054125、0.048、0.05025、0.052375,        0.04975、0.054125、0.055625、0.047125、0.0475、0.047,        0.049875、0.05025、0.048375、0.047、0.050625、0.05,        0.046625、0.04925、0.048、0.049125、0.05375、0.0545,        0.04925,0.049125,0.049125,0.049625,0.047,0.047625,        0.0535,0.051875,0.05075,0.04975,0.047375,0.049,        0.0485、0.050125、0.048、0.05475、0.05175、0.050125,        0.04725、0.0575、0.056875、0.047、0.0485、0.055375,        0.04975、0.047、0.0495、0.051375、0.04675、0.04925,        0.052125、0.04825、0.048125、0.046875、0.047、0.048625,        0.050875,0.05125,0.04825,0.052125,0.052375,0.05125,        0.049875、0.048625、0.04825、0.0475、0.048375、0.050875,        0.052875,0.0475,0.0485,0.05225,0.053625,0.05075,        0.0525,0.047125,0.0485,0.048875,0.049,0.0515,        0.055875、0.0515、0.05025、0.05125、0.054625、0.05525,        0.047、0.0545、0.052375、0.049875、0.051、0.048625,        0.0475,0.048,0.048875,0.050625,0.05375,0.051875,        0.048125、0.052125、0.048125、0.051、0.052625、0.048375,        0.047625、0.05、0.048125、0.050375、0.049125、0.053125,        0.053875、0.05075、0.052375、0.048875、0.05325、0.05825,        0.055625、0.0465、0.05475、0.051125、0.048375、0.0505,        0.04675,0.0495,0.04725,0.046625,0.049625,0.054,        0.056125、0.05175、0.050625、0.050375、0.047875、0.047,        0.048125、0.048875、0.050625、0.049875、0.047、0.0505,        0.047,0.053125,0.047625,0.05025,0.04825,0.05275,        0.051625,0.05,0.051625,0.05425,0.052,0.04775,        0.047,0.049125,0.05375,0.0535,0.04925,0.05125,        0.046375、0.04775、0.04775、0.0465、0.047、0.04675,        0.04675,0.04925,0.05125,0.046375,0.04825,0.0525,        0.057875、0.056375、0.054375、0.04825,0.0535,0.05475,        0.0485、0.048875、0.048625、0.0485、0.047625、0.046875,        0.0465、0.05125、0.054、0.05、0.048、0.047875,        0.0515,0.048125、0.055875、0.054875、0.051625、0.048125,        0.047625、0.048375、0.052875、0.0485、0.0475、0.0495,        0.05025、0.05675、0.0585、0.051625、0.05625、0.0605,        0.052125、0.0495、0.049、0.047875、0.051375、0.054125,        0.0525,0.0515,0.057875,0.055,0.05375,0.046375,        0.04775、0.0485、0.050125、0.050875、0.04925、0.049125,        0.0465、0.04975、0.053375、0.05225、0.0475、0.046375,        0.05375,0.049875,0.049875,0.047375,0.049125,0.049375,        0.04875、0.048125、0.05075、0.0505、0.046375、0.047375,        0.048625、0.0485、0.047125、0.052625、0.051125、0.04725,        0.050875、0.053875、0.0475、0.0495、0.051、0.055,        0.053,0.050125,0.04675,0.05375,0.054375,0.04725        0.046875、0.04925、0.04725、0.0495、0.05075、0.050875,        0.04775、0.05125、0.050125、0.047875、0.04825、0.046625,        0.0475,0.046375,0.04775,0.05075,0.048125,0.046375,        0.049625、0.0495、0.04675、0.046625、0.0475、0.04825,        0.053,0.050875,0.049,0.057875,0.058875,0.049875,        0.049125、0.0475、0.05225、0.055、0.055375、0.053875,        0.051125,0.049875,0.05025,0.050875,0.049,0.0575,        0.051875,0.049375,0.04775,0.051125,0.050375,0.0465,        0.047375、0.0465、0.046375、0.048875、0.051875、0.047,        0.047125、0.047125、0.046875、0.049625、0.048625、0.051,        0.049,0.046375,0.049,0.056125,0.054625,0.047625,        0.046625、0.0475、0.051875、0.05175、0.047625、0.050375,        0.055125、0.05275、0.047125、0.05325、0.060125、0.056625,        0.053,0.052125,0.047125,0.04825,0.050375,0.05025,        0.048,0.046625,0.047125,0.04875,0.047,0.05525,        0.0535,0.047,0.0495,0.0535,0.05125,0.046625,        0.0495,0.04675,0.04875,0.047125,0.04975,0.047,        0.049875、0.046875、0.047125、0.048、0.046375、0.0495,        0.04975、0.05125、0.048375、0.049125、0.0515、0.048375,        0.052375、0.051125、0.046375、0.047125、0.050375、0.0465,        0.052375、0.05375、0.04925、0.05025、0.0565、0.054875,        0.048,0.049375,0.052625,0.055375,0.053375,0.05075,        0.048875、0.05475、0.05075、0.0485、0.049125、0.0475,        0.047375、0.047375、0.047、0.052125、0.053875、0.049,        0.052625、0.0485、0.04675、0.04875、0.05、0.0545,        0.05025、0.0495、0.0515、0.0485、0.05025、0.0465,        0.0465、0.048375、0.06375、0.10175、0.11975、0.118375,        0.121375,0.12675,0.123,0.095375,0.055,0.05525,        0.04775、0.053125、0.052375、0.056625、0.0565、0.046875,        0.048、0.05175、0.048、0.052、0.048、0.048,        0.05175,0.05025,0.049625,0.049625,0.047375,0.046625,        0.052375、0.0555、0.051375、0.050625、0.052375、0.050125,        0.048、0.052125、0.052125、0.0495、0.048875、0.048,        0.049875、0.051125、0.050625、0.048、0.0465、0.048,        0.04675、0.050875、0.048、0.046625、0.0495、0.050375,        0.046625、0.0515、0.049875、0.049625、0.04675、0.049125,        0.05025、0.050375、0.04725、0.047625、0.047、0.051625,        0.0485、0.05225、0.046875、0.0475、0.04825、0.050375,        0.05725,0.052375,0.048,0.046375,0.0475,0.0495,        0.047875、0.046375、0.049875、0.046875、0.048、0.046875,        0.048625、0.047125、0.046625、0.05、0.048875、0.04675,        0.050125、0.05425、0.051375、0.050125、0.053375、0.052,        0.053875、0.048、0.05575、0.049875、0.052125、0.048875,        0.047375、0.048875、0.049125、0.047375、0.047375、0.047625,        0.0495,0.04825,0.047875,0.04875,0.054,0.052125,        0.051、0.046625、0.04925、0.05075、0.054375、0.0555,        0.051625、0.046625、0.052125、0.055875、0.047、0.053875,        0.050875、0.0505、0.0465、0.053125、0.050875、0.050625,        0.051125、0.050875、0.056875、0.04925、0.050625、0.054125,        0.056625、0.05025、0.0465、0.04675、0.049625、0.047,        0.048375、0.047125、0.04875、0.048375、0.048875、0.04775,        0.04775,0.047,0.052125,0.050875,0.054,0.058375,        0.054,0.049125,0.04675,0.051875,0.05425,0.050125,        0.04675,0.047625,0.046375,0.05275,0.053,0.04875,        0.049125、0.047125、0.049375、0.0475、0.051125、0.0495,        0.052375,0.047,0.047125,0.050875])


  [1]: https://i.imgur.com/OeKzvrb.png
  [2]: https://i.imgur.com/ALtba5F.png

1 个答案:

答案 0 :(得分:1)

问题1:

使用specshow时需要指定采样率:

librosa.display.specshow(stft, x_axis='time', y_axis='log', sr=sr)

否则,将使用默认值(22,050 Hz)(请参见docs)。

问题2:

librosa.core.frames_to_time不以stft[0]作为参数,这将是第一帧的频点。相反,它以帧数作为第一个参数。

想象一下,您有一个sr=10000 Hz的音频信号。然后,使用n_fft=2000hop_length=1000在其上运行STFT。然后,您每跳获得一个 frame ,并且该跳的长度为0.1s,因为10000个样本对应于1s,而1000个样本(1个跃点)因此对应于0.1s。

stft[0]不是帧号。相反,第一个stft的形状为(1 + n_fft/2, t)(请参见here)。这意味着第一维是频点,第二维是帧号(t)。

因此stft中的帧总数为stft.shape[1]。 要获取源音频的长度,可以执行以下操作:

time = librosa.core.frames_to_time(stft.shape[1], sr=sr, hop_length=hop_length, n_fft=n_fft)