Spaces:
No application file
No application file
commit
Browse files- extract_features.py +33 -0
extract_features.py
ADDED
@@ -0,0 +1,33 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import numpy as np
|
2 |
+
import soundfile
|
3 |
+
import librosa
|
4 |
+
|
5 |
+
def extract_feature(file_name, **kwargs):
|
6 |
+
|
7 |
+
chroma = kwargs.get("chroma")
|
8 |
+
contrast = kwargs.get("contrast")
|
9 |
+
mfcc = kwargs.get("mfcc")
|
10 |
+
mel = kwargs.get("mel")
|
11 |
+
tonnetz = kwargs.get("tonnetz")
|
12 |
+
|
13 |
+
with soundfile.SoundFile(file_name) as audio_clip:
|
14 |
+
X = audio_clip.read(dtype="float32")
|
15 |
+
sound_fourier = np.abs(librosa.stft(X)) # Conducting short time fourier transform of audio clip
|
16 |
+
result = np.array([])
|
17 |
+
|
18 |
+
if mfcc:
|
19 |
+
mfccs = np.mean(librosa.feature.mfcc(y=X, sr=audio_clip.samplerate, n_mfcc=40).T, axis=0)
|
20 |
+
result = np.hstack((result, mfccs))
|
21 |
+
if chroma:
|
22 |
+
chroma = np.mean(librosa.feature.chroma_stft(S=sound_fourier, sr=audio_clip.samplerate).T, axis=0)
|
23 |
+
result = np.hstack((result, chroma))
|
24 |
+
if mel:
|
25 |
+
mel = np.mean(librosa.feature.melspectrogram(X, sr=audio_clip.samplerate).T, axis=0)
|
26 |
+
result = np.hstack((result, mel))
|
27 |
+
if contrast:
|
28 |
+
contrast = np.mean(librosa.feature.spectral_contrast(S=sound_fourier, sr=audio_clip.samplerate).T, axis=0)
|
29 |
+
result = np.hstack((result, contrast))
|
30 |
+
if tonnetz:
|
31 |
+
tonnetz = np.mean(librosa.feature.tonnetz(y=librosa.effects.harmonic(X), sr=audio_clip.samplerate).T, axis=0)
|
32 |
+
result = np.hstack((result, tonnetz))
|
33 |
+
return result
|