Update app.py

Try the model trained by sanchit-gandhi

app.py

@@ -13,9 +13,11 @@ 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")
-processor = SpeechT5Processor.from_pretrained("kfahn/speecht5_finetuned_voxpopuli_cs")
+#processor = SpeechT5Processor.from_pretrained("kfahn/speecht5_finetuned_voxpopuli_cs")
+processor = SpeechT5Processor.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl")
 
-model = SpeechT5ForTextToSpeech.from_pretrained("kfahn/speecht5_finetuned_voxpopuli_cs").to(device)
+model = SpeechT5ForTextToSpeech.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl").to(device)
+#model = SpeechT5ForTextToSpeech.from_pretrained("kfahn/speecht5_finetuned_voxpopuli_cs").to(device)
 #model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
 vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
 
@@ -24,7 +26,7 @@ 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": "cs"})
+    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "nl"})
     return outputs["text"]