Update app.py

app.py

@@ -9,20 +9,20 @@ from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Proce
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
 # load speech translation checkpoint
-asr_pipe = pipeline("automatic-speech-recognition", model="jarguello76/whisper-tiny-dv", device=device)
+asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)
 
 # load text-to-speech checkpoint and speaker embeddings
 processor = SpeechT5Processor.from_pretrained("jarguello76/text-to-speech-speecht5_finetuned_voxpopuli_es")
 
 model = SpeechT5ForTextToSpeech.from_pretrained("jarguello76/text-to-speech-speecht5_finetuned_voxpopuli_es").to(device)
-vocoder = SpeechT5HifiGan.from_pretrained("jarguello76/text-to-speech-speecht5_finetuned_voxpopuli_es").to(device)
+vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
 
 embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
 speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
 
 
 def translate(audio):
-    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
+    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "es"})
     return outputs["text"]