Update app.py

app.py

@@ -24,36 +24,29 @@ def translate_speech(audio):
     print(f"Output: {output}")  # Print the output to see what it contains
 
     # Check if the output contains 'text'
-    if 'text' in output:
-        transcription = output["text"]
+    if 'text' in output[0]:
+        transcription = output[0]["text"]
     else:
         print("The output does not contain 'text'")
         return
 
     # Use the translation pipeline to translate the transcription
-    translated_text = translator(transcription, return_tensors="pt")
+    translated_text = translator(transcription)
     print(f"Translated text: {translated_text}")  # Print the translated text to see what it contains
 
-    # 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'])
-    else:
-        print("The translated text does not contain 'generated_token_ids'")
-        return
-
     # Use the VITS model to synthesize the translated text into speech
-    inputs = tokenizer(translated_text_str, return_tensors="pt")
+    inputs = tokenizer(translated_text[0]['translation_text'], return_tensors="pt")
     with torch.no_grad():
-        output = model(**inputs).waveform
+        output = model.generate(**inputs)
 
     # Save the synthesized speech to a WAV file
     scipy.io.wavfile.write("synthesized_speech.wav", rate=model.config.sampling_rate, data=output.float().numpy())
 
-    # Scale the audio data to the range of int16 format
-    synthesised_speech = (output * 32767).astype(np.int16)
+    print("Translated text:", translated_text[0]['translation_text'])
+    print("Synthesized speech data shape:", output.shape)
+    print("Sampling rate:", model.config.sampling_rate)
 
-    return 16000, synthesised_speech
+    return 16000, output.numpy()
 
 # Define the Gradio interface
 iface = gr.Interface(