Update app.py

app.py

@@ -5,33 +5,30 @@ from datasets import load_dataset
 
 from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
 
-
 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)
 
-# load text-to-speech checkpoint and speaker embeddings
+translate_pipeline = pipeline("translation_en_to_fr", model="t5-base", device=device)
+
 processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
 
-model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
+model = SpeechT5ForTextToSpeech.from_pretrained("Liphos/speecht5_tts_voxpopuli_fr").to(device)
 vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").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"})
-    return outputs["text"]
-
+    en_outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
+    fr_outputs = translate_pipeline(en_outputs["text"])
+    return fr_outputs[0]["translation_text"]
 
 def synthesise(text):
     inputs = processor(text=text, return_tensors="pt")
     speech = model.generate_speech(inputs["input_ids"].to(device), speaker_embeddings.to(device), vocoder=vocoder)
     return speech.cpu()
 
-
 def speech_to_speech_translation(audio):
     translated_text = translate(audio)
     synthesised_speech = synthesise(translated_text)
@@ -41,17 +38,13 @@ 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 French.
 """
-
 demo = gr.Blocks()
 
 mic_translate = gr.Interface(
     fn=speech_to_speech_translation,
-    inputs=gr.Audio(source="microphone", type="filepath"),
+    inputs=gr.Audio(sources="microphone", type="filepath"),
     outputs=gr.Audio(label="Generated Speech", type="numpy"),
     title=title,
     description=description,
@@ -59,9 +52,8 @@ mic_translate = gr.Interface(
 
 file_translate = gr.Interface(
     fn=speech_to_speech_translation,
-    inputs=gr.Audio(source="upload", type="filepath"),
+    inputs=gr.Audio(sources="upload", type="filepath"),
     outputs=gr.Audio(label="Generated Speech", type="numpy"),
-    examples=[["./example.wav"]],
     title=title,
     description=description,
 )
@@ -69,4 +61,4 @@ file_translate = gr.Interface(
 with demo:
     gr.TabbedInterface([mic_translate, file_translate], ["Microphone", "Audio File"])
 
-demo.launch()
+demo.launch()