BilalHasan
commited on
Commit
•
dac34e3
1
Parent(s):
3548181
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,44 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers import pipeline
|
2 |
+
from sessions import sessions
|
3 |
+
import torchaudio
|
4 |
+
import torchaudio.transforms as T
|
5 |
+
import gradio as gr
|
6 |
+
|
7 |
+
pipe = pipeline(
|
8 |
+
"audio-classification",
|
9 |
+
model="BilalHasan/distilhubert-finetuned-ravdess",
|
10 |
+
)
|
11 |
+
|
12 |
+
audio_batch = []
|
13 |
+
def split_audio(array):
|
14 |
+
len_of_each_array = 30 * 16000
|
15 |
+
arr1, arr2 = array[0: len_of_each_array], array[int(len_of_each_array / 2):]
|
16 |
+
audio_batch.append(arr1)
|
17 |
+
if len(arr2) > len_of_each_array:
|
18 |
+
split_audio(arr2)
|
19 |
+
else:
|
20 |
+
audio_batch.append(arr2)
|
21 |
+
return audio_batch
|
22 |
+
|
23 |
+
|
24 |
+
def prediction(path):
|
25 |
+
predictions = []
|
26 |
+
array, sr = torchaudio.load(path)
|
27 |
+
resampler = T.Resample(sr, 16000)
|
28 |
+
resampled_audio = resampler(array)
|
29 |
+
audio_batch = split_audio(resampled_audio[0].numpy())
|
30 |
+
for i in range(len(audio_batch)):
|
31 |
+
predictions.append(pipe(audio_batch[i])[0]['label'])
|
32 |
+
mood = max(set(predictions), key = predictions.count)
|
33 |
+
if mood in ['neutral', 'calm', 'happy', 'surprised']:
|
34 |
+
mood = 'other'
|
35 |
+
session = sessions.mood2session[mood]
|
36 |
+
return mood, session
|
37 |
+
|
38 |
+
|
39 |
+
demo = gr.Interface(
|
40 |
+
fn=prediction,
|
41 |
+
inputs=[gr.Audio(type="filepath")],
|
42 |
+
outputs=[gr.Textbox(label="Mood"), gr.Textbox(label="Session")]
|
43 |
+
)
|
44 |
+
demo.launch()
|