fix: enhance audio processing in transcribe function with librosa and improve transcription output

app.py

@@ -23,12 +23,13 @@ def load_model():
         print(f"Model loaded on device: {model.device}")
     return model
 
-
 @spaces.GPU(duration=120)
 def transcribe(audio, state=""):
     # Load the model inside the GPU worker process
     import numpy as np
     import soundfile as sf
+    import librosa
+    import os
     model = load_model()
     
     if audio is None or isinstance(audio, int):
@@ -36,17 +37,13 @@ def transcribe(audio, state=""):
         return state, state
     print(f"Received audio input of type: {type(audio)}")
     print(f"Audio shape: {audio.shape if isinstance(audio, np.ndarray) else 'N/A'}")
-    # Append NumPy array to buffer
-    # if isinstance(audio, tuple):
-    #     print(f"Tuple contents: {audio}")
-    #     # Try extracting the first element
-    #     audio = audio[1] if len(audio) > 1 else None
-        
+    
     if isinstance(audio, tuple) and len(audio) == 2 and isinstance(audio[1], np.ndarray):
         # Handle tuple of (sample_rate, audio_array)
         print(f"Tuple contents: {audio}")
         sample_rate, audio_data = audio
-        try:            # Resample to 16kHz for NeMo
+        try:
+            # Resample to 16kHz for NeMo
             if sample_rate != 16000:
                 print(f"Resampling from {sample_rate}Hz to 16000Hz")
                 audio_data = librosa.resample(audio_data.astype(float), orig_sr=sample_rate, target_sr=16000)
@@ -55,17 +52,20 @@ def transcribe(audio, state=""):
             sf.write(temp_file, audio_data, samplerate=16000)
             print(f"Processing temporary audio file: {temp_file}")
             
-            transcription = model.transcribe([temp_file])[0]
-            print(type(transcription))
+            # Transcribe and extract only the text (string)
+            hypothesis = model.transcribe([temp_file])[0]
+            print(f"Hypothesis: {hypothesis}")
+            transcription = hypothesis.text  # Extract the text attribute (string)
+            print(f"Transcription: {transcription}")
                     
             os.remove(temp_file)  # Clean up
             print("Temporary file removed.")
         except Exception as e:
             print(f"Error processing audio: {e}")
-            # return state, state
+            return state, state
 
         new_state = state + " " + transcription if state else transcription
-        print(new_state)
+        print(f"New state: {new_state}")
         return new_state, new_state
     return state, state