Update app.py

app.py

@@ -12,9 +12,10 @@ device = "cuda:0" if torch.cuda.is_available() else "cpu"
 asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)
 
 # load text-to-speech checkpoint and speaker embeddings
-processor = SpeechT5Processor.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl")
+model_name = "wuula/speecht5_tts_common_voice_zh"
+processor = SpeechT5Processor.from_pretrained("model_name")
 
-model = SpeechT5ForTextToSpeech.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl").to(device)
+model = SpeechT5ForTextToSpeech.from_pretrained("model_name").to(device)
 vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
 
 embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
@@ -22,8 +23,8 @@ speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze
 
 
 def translate(audio):
-    # outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "nl"})
-    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe"})
+    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "zh"})
+    # outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe"})
     return outputs["text"]