Update app.py

changed to multilingual dutch target language instead of english to convert input speech to text
changed to use the RajkNakka/speecht5_finetuned_voxpopuli_nl for synthesizing text to speech in dutch
only updated the

app.py

@@ -12,17 +12,18 @@ 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("microsoft/speecht5_tts")
+# processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
+processor = SpeechT5Processor.from_pretrained("RajkNakka/speecht5_finetuned_voxpopuli_nl")
 
-model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
-vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
+model = SpeechT5ForTextToSpeech.from_pretrained("RajkNakka/speecht5_finetuned_voxpopuli_nl").to(device)
+vocoder = SpeechT5HifiGan.from_pretrained("RajkNakka/speecht5_finetuned_voxpopuli_nl").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": "nl"})
     return outputs["text"]