Update speech translation checkpoint and text-to-speech checkpoint

app.py

@@ -9,12 +9,13 @@ from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Proce
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
 # load speech translation checkpoint
-asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)
+asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-tiny", device=device)
 
 # load text-to-speech checkpoint and speaker embeddings
-processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
+tts_checkpoint = "sanchit-gandhi/speecht5_tts_vox_nl"
+processor = SpeechT5Processor.from_pretrained(tts_checkpoint)
 
-model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
+model = SpeechT5ForTextToSpeech.from_pretrained(tts_checkpoint).to(device)
 vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
 
 embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
@@ -41,8 +42,8 @@ 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:
+Demo for cascaded speech-to-speech translation (STST), mapping from source speech in any language to target speech in Netherland. Demo uses OpenAI's [Whisper Tiny](https://huggingface.co/openai/whisper-tiny) model for speech translation, and a finetuned
+[SpeechT5 TTS](https://huggingface.co/sanchit-gandhi/speecht5_tts_vox_nl) 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")
 """