Updated to transcribe in catalan :o and README

app.py

@@ -27,7 +27,7 @@ speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze
 
 
 def translate(audio):
-    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate", "language": "catalan"})
+    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "catalan"})
     return outputs["text"]
 
 
@@ -44,10 +44,10 @@ def speech_to_speech_translation(audio):
     return 16000, synthesised_speech
 
 
-title = "Cascaded STST"
+title = "Demo STST - Multilingual to Català Speech"
 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 Català. Demo uses OpenAI's [Whisper Base](https://huggingface.co/openai/whisper-base) model for speech translation to català, and Microsoft's
+[SpeechT5 TTS](https://huggingface.co/microsoft/speecht5_tts) model for text-to-speech fine-tuned on [projecte-aina/openslr-slr69-ca-trimmed-denoised](https://huggingface.co/datasets/projecte-aina/openslr-slr69-ca-trimmed-denoised). This demo can be improve updating it with [projecte-aina/tts-ca-coqui-vits-multispeaker](https://huggingface.co/projecte-aina/tts-ca-coqui-vits-multispeaker) model:
 
 ![Cascaded STST](https://huggingface.co/datasets/huggingface-course/audio-course-images/resolve/main/s2st_cascaded.png "Diagram of cascaded speech to speech translation")
 """