Update app.py
Browse files
app.py
CHANGED
@@ -9,24 +9,17 @@ import tempfile
|
|
9 |
learn = load_learner('model.pkl')
|
10 |
labels = learn.dls.vocab
|
11 |
|
12 |
-
def audio_to_spectrogram(
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
else:
|
21 |
-
y, sr = librosa.load(audio, sr=None)
|
22 |
-
else: # Recorded audio
|
23 |
-
y, sr = librosa.load(audio, sr=None)
|
24 |
|
25 |
-
# Generate mel spectrogram
|
26 |
S = librosa.feature.melspectrogram(y=y, sr=sr, n_mels=128, fmax=8000)
|
27 |
S_dB = librosa.power_to_db(S, ref=np.max)
|
28 |
-
|
29 |
-
# Create and save spectrogram image
|
30 |
fig, ax = plt.subplots()
|
31 |
img = librosa.display.specshow(S_dB, x_axis='time', y_axis='mel', sr=sr, fmax=8000, ax=ax)
|
32 |
fig.colorbar(img, ax=ax, format='%+2.0f dB')
|
@@ -34,23 +27,20 @@ def audio_to_spectrogram(audio):
|
|
34 |
spectrogram_file = "spectrogram.png"
|
35 |
plt.savefig(spectrogram_file)
|
36 |
plt.close()
|
37 |
-
|
38 |
return spectrogram_file
|
39 |
|
40 |
def predict(audio):
|
41 |
spectrogram_file = audio_to_spectrogram(audio)
|
42 |
img = PILImage.create(spectrogram_file)
|
43 |
-
img = img.resize((512, 512))
|
44 |
pred, pred_idx, probs = learn.predict(img)
|
45 |
return {labels[i]: float(probs[i]) for i in range(len(labels))}
|
46 |
|
47 |
-
|
48 |
-
|
49 |
gr.Interface(
|
50 |
fn=predict,
|
51 |
-
inputs=
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
outputs=gr.components.Label(num_top_classes=3)
|
56 |
-
).launch()
|
|
|
9 |
learn = load_learner('model.pkl')
|
10 |
labels = learn.dls.vocab
|
11 |
|
12 |
+
def audio_to_spectrogram(audio_file):
|
13 |
+
if audio_file.endswith('.mp3'):
|
14 |
+
with tempfile.NamedTemporaryFile(suffix='.wav') as temp_wav:
|
15 |
+
audio = AudioSegment.from_mp3(audio_file)
|
16 |
+
audio.export(temp_wav.name, format='wav')
|
17 |
+
y, sr = librosa.load(temp_wav.name, sr=None)
|
18 |
+
else:
|
19 |
+
y, sr = librosa.load(audio_file, sr=None)
|
|
|
|
|
|
|
|
|
20 |
|
|
|
21 |
S = librosa.feature.melspectrogram(y=y, sr=sr, n_mels=128, fmax=8000)
|
22 |
S_dB = librosa.power_to_db(S, ref=np.max)
|
|
|
|
|
23 |
fig, ax = plt.subplots()
|
24 |
img = librosa.display.specshow(S_dB, x_axis='time', y_axis='mel', sr=sr, fmax=8000, ax=ax)
|
25 |
fig.colorbar(img, ax=ax, format='%+2.0f dB')
|
|
|
27 |
spectrogram_file = "spectrogram.png"
|
28 |
plt.savefig(spectrogram_file)
|
29 |
plt.close()
|
|
|
30 |
return spectrogram_file
|
31 |
|
32 |
def predict(audio):
|
33 |
spectrogram_file = audio_to_spectrogram(audio)
|
34 |
img = PILImage.create(spectrogram_file)
|
35 |
+
img = img.resize((512, 512))
|
36 |
pred, pred_idx, probs = learn.predict(img)
|
37 |
return {labels[i]: float(probs[i]) for i in range(len(labels))}
|
38 |
|
39 |
+
examples = ['example_audio.mp3']
|
40 |
+
|
41 |
gr.Interface(
|
42 |
fn=predict,
|
43 |
+
inputs=gr.Audio(sources="upload", type="filepath", label="Upload audio (WAV or MP3)"),
|
44 |
+
outputs=gr.components.Label(num_top_classes=3),
|
45 |
+
examples=examples,
|
46 |
+
).launch()
|
|
|
|