Initial commit

app.py

@@ -8,46 +8,36 @@ import numpy as np
 @spaces.GPU(duration=60)
 def transcribe_and_respond(audio_file):
     try:
-        pipe = transformers.pipeline(
-            model='sarvamai/shuka_v1',
-            trust_remote_code=True,
-            device=0,
-            torch_dtype=torch.bfloat16
-        )
-
-        # Load the audio file
+        # Load audio
         audio, sr = librosa.load(audio_file, sr=16000)
-
-        # Print audio properties for debugging
         print(f"Audio dtype: {audio.dtype}, Audio shape: {audio.shape}, Sample rate: {sr}")
 
-        turns = [
-            {'role': 'system', 'content': 'Respond naturally and informatively.'},
-            {'role': 'user', 'content': '<|audio|>'}
-        ]
-
-        # Debug: Print the initial turns
-        print(f"Initial turns: {turns}")
+        # Initialize the model pipeline with an appropriate task
+        pipe = transformers.pipeline(
+            task="automatic-speech-recognition",  # Change to "text2text-generation" if needed
+            model="sarvamai/shuka_v1",
+            trust_remote_code=True,
+            device=0
+        )
 
-        # Call the model with the audio and prompt
-        output = pipe({'audio': audio, 'turns': turns, 'sampling_rate': sr}, max_new_tokens=512)
+        # Pass raw audio for transcription
+        output = pipe(audio_file)
 
-        # Debug: Print the final output from the model
         print(f"Model output: {output}")
-
-        return output
+        return output["text"] if isinstance(output, dict) and "text" in output else str(output)
 
     except Exception as e:
         return f"Error: {str(e)}"
 
+# Gradio interface
 iface = gr.Interface(
     fn=transcribe_and_respond,
     inputs=gr.Audio(sources="microphone", type="filepath"),
     outputs="text",
-    title="Live Transcription and Response",
-    description="Speak into your microphone, and the model will respond naturally and informatively.",
+    title="Live Transcription with Shuka v1",
+    description="Speak into your microphone, and the model will transcribe or respond using SarvamAI's Shuka v1.",
     live=True
 )
 
 if __name__ == "__main__":
-    iface.launch()
+    iface.launch(share=True)