Update app.py

app.py

@@ -18,17 +18,17 @@ tokenizer = WhisperTokenizer.from_pretrained("openai/whisper-base", language="fr
 whisper_model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-base")
 forced_decoder_ids = tokenizer.get_decoder_prompt_ids(language="french", task="translate")
 
-# load text-to-speech checkpoint and speaker embeddings
-# processor = SpeechT5Processor.from_pretrained("leo-kwan/speecht5_finetuned_voxpopuli_lt")
+load text-to-speech checkpoint and speaker embeddings
+processor = SpeechT5Processor.from_pretrained("leo-kwan/speecht5_finetuned_voxpopuli_lt")
 
-# model = SpeechT5ForTextToSpeech.from_pretrained("leo-kwan/speecht5_finetuned_voxpopuli_lt").to(device)
-# vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
+model = SpeechT5ForTextToSpeech.from_pretrained("leo-kwan/speecht5_finetuned_voxpopuli_lt").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)
+embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
+speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
 
-bark_model = BarkModel.from_pretrained("suno/bark-small")
-processor = BarkProcessor.from_pretrained("suno/bark-small")
+# bark_model = BarkModel.from_pretrained("suno/bark-small")
+# processor = BarkProcessor.from_pretrained("suno/bark-small")
 
 def translate(audio):
     # load speech translation checkpoint
@@ -45,18 +45,18 @@ def translate(audio):
 
 def synthesise(text):
     
-    inputs = processor(text, voice_preset="v2/fr_speaker_1")
-    speech = bark_model.generate(**inputs).cpu()
+    # inputs = processor(text, voice_preset="v2/fr_speaker_1")
+    # speech = bark_model.generate(**inputs).cpu()
     
-    # inputs = processor(text=text, return_tensors="pt")
-    # speech = model.generate_speech(inputs["input_ids"].to(device), speaker_embeddings.to(device), vocoder=vocoder)
+    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)
-    # synthesised_speech = (synthesised_speech.numpy() * 32767).astype(np.int16)
+    synthesised_speech = (synthesised_speech.numpy() * 32767).astype(np.int16)
     return 16000, synthesised_speech