Update app.py

Trying the French translation.

app.py

@@ -3,46 +3,35 @@ import numpy as np
 import torch
 from datasets import load_dataset
 
-from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
-
-from transliterate import translit
+from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline, VitsModel, VitsTokenizer
 
 
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
-# загрузить контрольную точку модели транскибации и перевода речи
+# load speech translation checkpoint
 asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)
-translate_pipe = pipeline("translation", model="Helsinki-NLP/opus-mt-en-ru")
-
-# Загрузим контрольную точку преобразования текста в речь и эбеддинги дикторов
-processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
-
-model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").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)
+# load text-to-speech checkpoint and speaker embeddings
+model = VitsModel.from_pretrained("Matthijs/mms-tts-fra")
+tokenizer = VitsTokenizer.from_pretrained("Matthijs/mms-tts-fra")
 
 
 def translate(audio):
-    transcription = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "english"})
-    translation = translate_pipe(transcription["text"])
-    result = translit(translation[0]['translation_text'], "ru", reversed=True)
-    print(result)
-    return result
+    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "french"})
+    return outputs["text"]
 
 
 def synthesise(text):
-    print(text)
-    inputs = processor(text=text, return_tensors="pt")
-    speech = model.generate_speech(inputs["input_ids"].to(device), speaker_embeddings.to(device), vocoder=vocoder)
+    inputs = tokenizer(text=text, return_tensors="pt")
+    speech_output = model(inputs["input_ids"].to(device))
+    speech = speech_output.audio[0]
     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.detach().numpy() * 32767).astype(np.int16)
     return 16000, synthesised_speech