Update app.py

app.py

@@ -1,52 +1,58 @@
+from __future__ import annotations
 import os
+import gradio as gr
 import torch
 import torchaudio
-import gradio as gr
+import spaces
 import nemo.collections.asr as nemo_asr
 
-# Select device
-device = "cuda:0" if torch.cuda.is_available() else "cpu"
+LANGUAGE_NAME_TO_CODE = {
+    "Assamese": "as", "Bengali": "bn", "Bodo": "br", "Dogri": "doi",
+    "Gujarati": "gu", "Hindi": "hi", "Kannada": "kn", "Kashmiri": "ks",
+    "Konkani": "kok", "Maithili": "mai", "Malayalam": "ml", "Manipuri": "mni",
+    "Marathi": "mr", "Nepali": "ne", "Odia": "or", "Punjabi": "pa",
+    "Sanskrit": "sa", "Santali": "sat", "Sindhi": "sd", "Tamil": "ta",
+    "Telugu": "te", "Urdu": "ur"
+}
+
+DESCRIPTION = """IndicConformer: Dual-Decoder ASR for Indian Languages"""
+
+device = "cuda:0" if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu"
+model = nemo_asr.models.EncDecCTCModel.from_pretrained("ai4bharat/IndicConformer").to(device)
+model.eval()
 
-# Load CTC and RNNT models from AI4Bharat
-asr_ctc = nemo_asr.models.EncDecCTCModelBPE.from_pretrained("ai4bharat/indicwhisper-ctc-indic").to(device)
-asr_rnnt = nemo_asr.models.EncDecRNNTBPEModel.from_pretrained("ai4bharat/indicwhisper-rnnt-indic").to(device)
+@spaces.GPU
+def transcribe_ctc_and_rnnt(audio_path, language_name):
+    lang_id = LANGUAGE_NAME_TO_CODE[language_name]
+    waveform, sample_rate = torchaudio.load(audio_path)
+    waveform = waveform.mean(dim=0, keepdim=True) if waveform.shape[0] > 1 else waveform
+    waveform = torchaudio.functional.resample(waveform, sample_rate, 16000)
+    waveform_np = waveform.squeeze().numpy()
 
-# All 22 scheduled Indian languages
-language_options = [
-    "Assamese", "Bengali", "Bodo", "Dogri", "Gujarati", "Hindi",
-    "Kannada", "Kashmiri", "Konkani", "Maithili", "Malayalam",
-    "Manipuri", "Marathi", "Nepali", "Odia", "Punjabi", "Sanskrit",
-    "Santali", "Sindhi", "Tamil", "Telugu", "Urdu"
-]
+    model.cur_decoder = "ctc"
+    ctc = model.transcribe([waveform_np], batch_size=1, language_id=lang_id)[0][0]
 
-# CTC ASR function
-def run_asr_ctc(audio_path, source_lang):
-    asr_ctc.change_vocabulary(language=source_lang)
-    return asr_ctc.transcribe(paths2audio_files=[audio_path])[0]
+    model.cur_decoder = "rnnt"
+    rnnt = model.transcribe([waveform_np], batch_size=1, language_id=lang_id)[0][0]
 
-# RNNT ASR function
-def run_asr_rnnt(audio_path, source_lang):
-    asr_rnnt.change_vocabulary(language=source_lang)
-    return asr_rnnt.transcribe(paths2audio_files=[audio_path])[0]
+    return ctc, rnnt
 
-# Gradio UI
 with gr.Blocks() as demo:
-    gr.Markdown("## AI4Bharat Indic ASR (CTC & RNNT)")
-    
-    with gr.Tab("CTC Transcription"):
-        with gr.Row():
-            input_audio = gr.Audio(type="filepath", label="Upload Audio")
-            source_lang = gr.Dropdown(choices=language_options, label="Language", value="Hindi")
-        output_text_ctc = gr.Textbox(label="CTC Transcription Output")
-        ctc_button = gr.Button("Transcribe (CTC)")
-        ctc_button.click(run_asr_ctc, inputs=[input_audio, source_lang], outputs=output_text_ctc)
-
-    with gr.Tab("RNNT Transcription"):
-        with gr.Row():
-            input_audio_rnnt = gr.Audio(type="filepath", label="Upload Audio")
-            source_lang_rnnt = gr.Dropdown(choices=language_options, label="Language", value="Hindi")
-        output_text_rnnt = gr.Textbox(label="RNNT Transcription Output")
-        rnnt_button = gr.Button("Transcribe (RNNT)")
-        rnnt_button.click(run_asr_rnnt, inputs=[input_audio_rnnt, source_lang_rnnt], outputs=output_text_rnnt)
-
-demo.launch()
+    gr.Markdown(DESCRIPTION)
+    with gr.Row():
+        with gr.Column():
+            audio = gr.Audio(label="Upload or record audio", type="filepath")
+            lang = gr.Dropdown(
+                label="Select language",
+                choices=LANGUAGE_NAME_TO_CODE.keys(),
+                value="Hindi"
+            )
+            transcribe_btn = gr.Button("Transcribe (CTC + RNNT)")
+        with gr.Column():
+            ctc_output = gr.Textbox(label="CTC Transcription")
+            rnnt_output = gr.Textbox(label="RNNT Transcription")
+
+    transcribe_btn.click(fn=transcribe_ctc_and_rnnt, inputs=[audio, lang], outputs=[ctc_output, rnnt_output])
+
+if __name__ == "__main__":
+    demo.queue().launch()