tahirjm commited on
Commit
61a93bc
Β·
verified Β·
1 Parent(s): fc83c2c

Create app.py

Browse files
Files changed (1) hide show
  1. app.py +49 -0
app.py ADDED
@@ -0,0 +1,49 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import gradio as gr
2
+ import soundfile as sf
3
+ import numpy as np
4
+ import tempfile
5
+ import torchaudio
6
+ from transformers import AutoModel
7
+
8
+ # Load ASR Model
9
+ def load_model():
10
+ return AutoModel.from_pretrained("ai4bharat/indic-conformer-600m-multilingual", trust_remote_code=True)
11
+
12
+ model = load_model()
13
+
14
+ def process_audio(audio, language, decoding_method):
15
+ if isinstance(audio, tuple): # Recorded audio
16
+ sample_rate, data = audio
17
+ temp_wav = tempfile.NamedTemporaryFile(delete=False, suffix=".wav")
18
+ sf.write(temp_wav.name, data, sample_rate)
19
+ audio_path = temp_wav.name
20
+ else: # Uploaded file
21
+ audio_path = audio
22
+
23
+ # Load and resample audio
24
+ wav, sr = torchaudio.load(audio_path)
25
+ target_sample_rate = 16000
26
+ if sr != target_sample_rate:
27
+ resampler = torchaudio.transforms.Resample(orig_freq=sr, new_freq=target_sample_rate)
28
+ wav = resampler(wav)
29
+
30
+ # Perform ASR with selected decoding method
31
+ transcription = model(wav, language, decoding_method)
32
+
33
+ return transcription
34
+
35
+ iface = gr.Interface(
36
+ fn=process_audio,
37
+ inputs=[
38
+ gr.Audio(source="microphone", type="numpy"),
39
+ gr.Audio(source="upload"),
40
+ gr.Dropdown(["hi", "ta", "bn", "mr", "te", "gu", "kn", "ml", "pa", "ur"], label="Select Language"),
41
+ gr.Radio(["ctc", "rnnt"], label="Decoding Method")
42
+ ],
43
+ outputs="text",
44
+ title="Multilingual ASR with Indic-Conformer",
45
+ description="Record or upload an audio file, select a language and decoding method, and transcribe it using the AI4Bharat Indic-Conformer model."
46
+ )
47
+
48
+ if __name__ == "__main__":
49
+ iface.launch()