Update app.py

app.py

@@ -11,7 +11,7 @@ device = "cuda:0" if torch.cuda.is_available() else "cpu"
 asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-large-v2", device=device)
 
 # load text-to-speech checkpoint and speaker embeddings
-model_id = "Sandiago21/speecht5_finetuned_facebook_voxpopuli_spanish"  # update with your model id
+model_id = "ckandemir/speecht5_finetuned_voxpopuli_fr"  # update with your model id
 # pipe = pipeline("automatic-speech-recognition", model=model_id)
 model = SpeechT5ForTextToSpeech.from_pretrained(model_id)
 vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")
@@ -21,44 +21,31 @@ speaker_embeddings = torch.tensor(embeddings_dataset[7440]["xvector"]).unsqueeze
 processor = SpeechT5Processor.from_pretrained(model_id)
 
 replacements = [
-     ("á", "a"),
-     ("ç", "c"),
-     ("è", "e"),
-     ("ì", "i"),
-     ("í", "i"),
-     ("ò", "o"),
-     ("ó", "o"),
-     ("ù", "u"),
-     ("ú", "u"),
-     ("š", "s"),
-     ("ï", "i"),
-     ("ñ", "n"),
-     ("ü", "u"),
+    ("à", "a"), ("â", "a"),
+    ("ç", "c"),
+    ("é", "e"), ("è", "e"), ("ê", "e"), ("ë", "e"),
+    ("î", "i"), ("ï", "i"),
+    ("ô", "o"),
+    ("ù", "u"), ("û", "u"),
 ]
 
+
 def cleanup_text(text):
     for src, dst in replacements:
         text = text.replace(src, dst)
     return text
 
-def synthesize_speech(text):
-    text = cleanup_text(text)
-    inputs = processor(text=text, return_tensors="pt")
-    speech = model.generate_speech(inputs["input_ids"].to(device), speaker_embeddings.to(device), vocoder=vocoder)
-
-    return gr.Audio.update(value=(16000, speech.cpu().numpy()))
-
 def translate(audio):
-    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "spanish"})
+    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "french"})
     return outputs["text"]
 
 
+
 def synthesise(text):
     text = cleanup_text(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()
-
+    return gr.Audio.update(value=(16000, speech.cpu().numpy()))
 
 def speech_to_speech_translation(audio):
     translated_text = translate(audio)
@@ -69,8 +56,8 @@ 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 Spanish. Demo uses OpenAI's [Whisper Large v2](https://huggingface.co/openai/whisper-large-v2) model for speech translation, and [Sandiago21/speecht5_finetuned_facebook_voxpopuli_spanish](https://huggingface.co/Sandiago21/speecht5_finetuned_facebook_voxpopuli_spanish) checkpoint for text-to-speech, which is based on Microsoft's
-[SpeechT5 TTS](https://huggingface.co/microsoft/speecht5_tts) model for text-to-speech, fine-tuned in Spanish Audio dataset:
+Demo for cascaded speech-to-speech translation (STST), mapping from source speech in any language to target speech in French. Demo uses OpenAI's [Whisper Large v2](https://huggingface.co/openai/whisper-large-v2) model for speech translation, and [ckandemir/speecht5_finetuned_voxpopuli_fr"](https://huggingface.co/ckandemir/speecht5_finetuned_voxpopuli_fr) checkpoint for text-to-speech, which is based on Microsoft's
+[SpeechT5 TTS](https://huggingface.co/microsoft/speecht5_tts) model for text-to-speech, fine-tuned in French Audio dataset:
 ![Cascaded STST](https://huggingface.co/datasets/huggingface-course/audio-course-images/resolve/main/s2st_cascaded.png "Diagram of cascaded speech to speech translation")
 """