Update app.py

app.py

@@ -14,11 +14,7 @@ voices = {
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
 # load speech translation checkpoint
-asr_pipe = pipeline("automatic-speech-recognition", model=ASR_MODEL_NAME, chunk_length_s=10,device=device)
-# set the forced ids
-#asr_pipe.model.config.forced_decoder_ids = asr_pipe.tokenizer.get_decoder_prompt_ids(language='de', task="transcribe")
-asr_pipe.config.forced_decoder_ids = None
-asr_pipe.config.suppress_tokens = []
+asr_pipe = pipeline("automatic-speech-recognition", model=ASR_MODEL_NAME, chunk_length_s=30,device=device)
 
 MULTILINGUAL = True  # set True for multilingual models, False for English-only
 
@@ -30,6 +26,10 @@ processor = AutoProcessor.from_pretrained("suno/bark-small")
 model = BarkModel.from_pretrained("suno/bark-small").to(device)
 sampling_rate = model.generation_config.sample_rate
 
+# set the forced ids
+model.config.forced_decoder_ids = None
+model.config.suppress_tokens = []
+
 def translate(audio):
     outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
     return outputs["text"]