Update app.py

app.py

@@ -25,7 +25,7 @@ speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze
 def translate(audio):
     # Trick Whisper to translate from any language to Dutch.
     # Note that using task=translate will translate to English instead.
-    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate", "language": "dutch"})
+    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "dutch"})
     return outputs["text"]
 
 
@@ -34,9 +34,10 @@ def synthesise(text):
     # Otherwise model.generate_speech will throw errors.
     inputs = processor(
         text=text,
+        # max_length=200,
         max_length=598,
         truncation=True,
-        padding=True,
+        # padding=True,
         return_tensors="pt"
     )
     # inputs = processor(text=text, return_tensors="pt")
@@ -57,7 +58,6 @@ Demo for cascaded speech-to-speech translation (STST), mapping from source speec
 [SpeechT5 TTS](https://huggingface.co/sanchit-gandhi/speecht5_tts_vox_nl) model for text-to-speech:
 ![Cascaded STST](https://huggingface.co/datasets/huggingface-course/audio-course-images/resolve/main/s2st_cascaded.png "Diagram of cascaded speech to speech translation")
 """
-
 demo = gr.Blocks()
 
 mic_translate = gr.Interface(