Update app.py

app.py

@@ -15,53 +15,53 @@ def translate_speech(audio_data_tuple):
     # Extract the audio data from the tuple
     sample_rate, audio_data = audio_data_tuple
 
-    # Save the audio data to a temporary file
-    with tempfile.NamedTemporaryFile(suffix=".wav", delete=True) as temp_audio_file:
-        sf.write(temp_audio_file.name, audio_data, sample_rate)
+    # Resample the audio data to 16000 Hz
+    audio_data_resampled = librosa.resample(audio_data, sample_rate, 16000)
 
-        # Prepare the input dictionary
-        input_dict = asr_processor(audio_data, sampling_rate=sample_rate, return_tensors="pt", padding=True)
+    # Prepare the input dictionary
+    input_dict = asr_processor(audio_data_resampled, sampling_rate=16000, return_tensors="pt", padding=True)  # Pass the resampled audio_data here
 
-        # Use the ASR model to get the logits
-        logits = asr_model(input_dict.input_values.to("cpu")).logits
+    # Use the ASR model to get the logits
+    logits = asr_model(input_dict.input_values.to("cpu")).logits
 
-        # Get the predicted IDs
-        pred_ids = torch.argmax(logits, dim=-1)[0]
+    # Get the predicted IDs
+    pred_ids = torch.argmax(logits, dim=-1)[0]
 
-        # Decode the predicted IDs to get the transcription
-        transcription = asr_processor.decode(pred_ids)
-        print(f"Transcription: {transcription}")  # Print the transcription
+    # Decode the predicted IDs to get the transcription
+    transcription = asr_processor.decode(pred_ids)
+    print(f"Transcription: {transcription}")  # Print the transcription
 
-        # Use the translation pipeline to translate the transcription
-        translated_text = translator(transcription, return_tensors="pt")
-        print(f"Translated text: {translated_text}")  # Print the translated text
+    # Use the translation pipeline to translate the transcription
+    translated_text = translator(transcription, return_tensors="pt")
+    print(f"Translated text: {translated_text}")  # Print the translated text
 
-        # Check if the translated text contains 'generated_token_ids'
-        if 'generated_token_ids' in translated_text[0]:
-            # Decode the tokens into text
-            translated_text_str = translator.tokenizer.decode(translated_text[0]['generated_token_ids'])
-            print(f"Translated text string: {translated_text_str}")  # Print the translated text string
-        else:
-            print("The translated text does not contain 'generated_token_ids'")
-            return
+    # Check if the translated text contains 'generated_token_ids'
+    if 'generated_token_ids' in translated_text[0]:
+        # Decode the tokens into text
+        translated_text_str = translator.tokenizer.decode(translated_text[0]['generated_token_ids'])
+        print(f"Translated text string: {translated_text_str}")  # Print the translated text string
+    else:
+        print("The translated text does not contain 'generated_token_ids'")
+        return
 
-        # Use the text-to-speech pipeline to synthesize the translated text
-        synthesised_speech = tts(translated_text_str)
+    # Use the text-to-speech pipeline to synthesize the translated text
+    synthesised_speech = tts(translated_text_str)
 
-        # Check if the synthesised speech contains 'audio'
-        if 'audio' in synthesised_speech:
-            synthesised_speech_data = synthesised_speech['audio']
-        else:
-            print("The synthesised speech does not contain 'audio'")
-            return
+    # Check if the synthesised speech contains 'audio'
+    if 'audio' in synthesised_speech:
+        synthesised_speech_data = synthesised_speech['audio']
+    else:
+        print("The synthesised speech does not contain 'audio'")
+        return
 
-        # Flatten the audio data
-        synthesised_speech_data = synthesised_speech_data.flatten()
+    # Flatten the audio data
+    synthesised_speech_data = synthesised_speech_data.flatten()
 
-        # Scale the audio data to the range of int16 format
-        synthesised_speech = (synthesised_speech_data * 32767).astype(np.int16)
+    # Scale the audio data to the range of int16 format
+    synthesised_speech = (synthesised_speech_data * 32767).astype(np.int16)
+
+    return 16000, synthesised_speech
 
-        return 16000, synthesised_speech
 
 # Define the Gradio interface
 iface = gr.Interface(