Update app.py

app.py

@@ -10,7 +10,6 @@ device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
 # load speech translation checkpoint
 asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)
-forced_decoder_ids = processor.get_decoder_prompt_ids(language="spanish", task="translate")
 
 
 # load text-to-speech checkpoint and speaker embeddings
@@ -27,6 +26,9 @@ def translate(audio):
     outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
     return outputs["text"]
 
+    set:"transcribe""language"
+    generate_kwargs={"task": "transcribe", "language": "es"}
+
 
 def synthesise(text):
     inputs = processor(text=text, return_tensors="pt")