Update app.py

app.py

@@ -3,7 +3,7 @@ import numpy as np
 import torch
 from datasets import load_dataset
 
-from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
+from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline, AutoProcessor
 
 
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
@@ -12,9 +12,12 @@ 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")
+#model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
+
+processor = AutoProcessor.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl")
+model = SpeechT5ForTextToSpeech.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl").to(device)
 
-model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
 vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
 
 embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
@@ -22,7 +25,7 @@ speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze
 
 
 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"]
 
 
@@ -41,10 +44,7 @@ def speech_to_speech_translation(audio):
 
 title = "Cascaded STST"
 description = """
-Demo for cascaded speech-to-speech translation (STST), mapping from source speech in any language to target speech in English. Demo uses OpenAI's [Whisper Base](https://huggingface.co/openai/whisper-base) model for speech translation, and Microsoft's
-[SpeechT5 TTS](https://huggingface.co/microsoft/speecht5_tts) model for text-to-speech:
-
-![Cascaded STST](https://huggingface.co/datasets/huggingface-course/audio-course-images/resolve/main/s2st_cascaded.png "Diagram of cascaded speech to speech translation")
+Demo for cascaded speech-to-speech translation (STST), mapping from source speech in any language to target speech in Dutch.
 """
 
 demo = gr.Blocks()