Update app.py

app.py

@@ -14,11 +14,13 @@ asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base",
 # load text-to-speech checkpoint and speaker embeddings
 #processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
 #Use own TTS Model
-processor = SpeechT5Processor.from_pretrained("jasonl1/speecht5_finetuned_voxpopuli_fi")
+#processor = SpeechT5Processor.from_pretrained("jasonl1/speecht5_finetuned_voxpopuli_fi")
+processor = SpeechT5Processor.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl")
 
 #model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
 #Use own TTS Model
-model = SpeechT5ForTextToSpeech.from_pretrained("jasonl1/speecht5_finetuned_voxpopuli_fi")
+'model = SpeechT5ForTextToSpeech.from_pretrained("jasonl1/speecht5_finetuned_voxpopuli_fi")
+model = SpeechT5ForTextToSpeech.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl")
 
 vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
 
@@ -35,7 +37,7 @@ speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze
 # At Inference. it should use translate(sample["audio"].copy())
 
 def translate(audio):
-    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "fi"})
+    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "nl"})
     return outputs["text"]