Spaces:
Running
Running
converting stereo audio to mono if needed
Browse files
app.py
CHANGED
@@ -1,4 +1,4 @@
|
|
1 |
-
|
2 |
|
3 |
import numpy as np
|
4 |
|
@@ -20,6 +20,12 @@ LM_HUB_FP = 'language_model/cv8be_5gram.bin'
|
|
20 |
def main(audio_fp: str):
|
21 |
audio, sampling_rate = torchaudio.load(audio_fp, normalize=True)
|
22 |
|
|
|
|
|
|
|
|
|
|
|
|
|
23 |
# resample audio to 16kHz
|
24 |
resampler = Resample(orig_freq=sampling_rate, new_freq=16_000)
|
25 |
audio_resampled = resampler(audio)
|
@@ -37,17 +43,20 @@ def main(audio_fp: str):
|
|
37 |
|
38 |
res['sampling_rate_orig'] = sampling_rate
|
39 |
res['init_audio_shape'] = audio.shape
|
|
|
40 |
res['inputs_shape'] = inputs.shape
|
41 |
res['inputs_max'] = np.max(inputs).item()
|
42 |
res['inputs_min'] = np.min(inputs).item()
|
43 |
|
44 |
-
|
|
|
|
|
45 |
|
46 |
|
47 |
iface = gr.Interface(
|
48 |
fn=main,
|
49 |
inputs=gr.inputs.Audio(
|
50 |
-
source='microphone', type='filepath',
|
51 |
label='Запішыце аўдыяфайл, каб распазнаваць маўленне'
|
52 |
),
|
53 |
outputs='text'
|
|
|
1 |
+
import json
|
2 |
|
3 |
import numpy as np
|
4 |
|
|
|
20 |
def main(audio_fp: str):
|
21 |
audio, sampling_rate = torchaudio.load(audio_fp, normalize=True)
|
22 |
|
23 |
+
# convert stereo to mono
|
24 |
+
converted_to_mono = False
|
25 |
+
if audio.shape[0] > 1:
|
26 |
+
audio = torch.mean(audio, dim=0, keepdim=True)
|
27 |
+
converted_to_mono = True
|
28 |
+
|
29 |
# resample audio to 16kHz
|
30 |
resampler = Resample(orig_freq=sampling_rate, new_freq=16_000)
|
31 |
audio_resampled = resampler(audio)
|
|
|
43 |
|
44 |
res['sampling_rate_orig'] = sampling_rate
|
45 |
res['init_audio_shape'] = audio.shape
|
46 |
+
res['converted_to_mono'] = converted_to_mono
|
47 |
res['inputs_shape'] = inputs.shape
|
48 |
res['inputs_max'] = np.max(inputs).item()
|
49 |
res['inputs_min'] = np.min(inputs).item()
|
50 |
|
51 |
+
res_str = json.dumps(res, indent=2)
|
52 |
+
|
53 |
+
return res_str
|
54 |
|
55 |
|
56 |
iface = gr.Interface(
|
57 |
fn=main,
|
58 |
inputs=gr.inputs.Audio(
|
59 |
+
source='microphone', type='filepath',
|
60 |
label='Запішыце аўдыяфайл, каб распазнаваць маўленне'
|
61 |
),
|
62 |
outputs='text'
|