Update app.py

app.py

@@ -1,3 +1,7 @@
+"""Pronkin_hw_task3.ipynb
+https://colab.research.google.com/drive/149j9u-wsD3GiEwRA8clBrXQ8bh5DRk7I?usp=sharing
+"""
+
 import gradio as gr
 import numpy as np
 import torch
@@ -8,43 +12,51 @@ from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Proce
 
 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)
+asr_pipe = pipeline("automatic-speech-recognition", model="voidful/wav2vec2-xlsr-multilingual-56", device=device)
+
+processor = WhisperProcessor.from_pretrained("openai/whisper-small")
+
+translator_1 = pipeline("translation", model="Helsinki-NLP/opus-mt-mul-en")
+translator_2 = pipeline("translation", model="Helsinki-NLP/opus-mt-en-ru")
 
-# load text-to-speech checkpoint and speaker embeddings
-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)
+model = VitsModel.from_pretrained("facebook/mms-tts-rus")
+tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-rus")
 
-embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
-speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
 
+def translator_mul_ru(text):
+
+    translation  = translator_2(translator_1(text)[0]['translation_text'])
+    return translation[0]['translation_text']
 
 def translate(audio):
     outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
     return outputs["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)
+    translated_text = translator_mul_ru(text)
+    inputs = tokenizer(translated_text, return_tensors="pt")
+    input_ids = inputs["input_ids"]
+
+    with torch.no_grad():
+        outputs = model(input_ids)
+    speech = outputs["waveform"]
     return speech.cpu()
 
 
 def speech_to_speech_translation(audio):
     translated_text = translate(audio)
+    print(translated_text)
     synthesised_speech = synthesise(translated_text)
     synthesised_speech = (synthesised_speech.numpy() * 32767).astype(np.int16)
-    return 16000, synthesised_speech
+    return 16000, synthesised_speech[0]
 
 
-title = "Cascaded STST"
+title = "Pronkin custom 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")
+* ASR-модель распознает речь с помощью voidful/wav2vec2-xlsr-multilingual-56 и возвращает текст на любом из 56 языков.
+* Перевод текста с любого на английский с помощью модели Helsinki-NLP/opus-mt-mul-en, с английского на русский - Helsinki-NLP/opus-mt-en-ru 
+* Синтез речи на русском языке с помощью модели https://huggingface.co/facebook/mms-tts-rus
 """
 
 demo = gr.Blocks()
@@ -61,12 +73,11 @@ file_translate = gr.Interface(
     fn=speech_to_speech_translation,
     inputs=gr.Audio(source="upload", type="filepath"),
     outputs=gr.Audio(label="Generated Speech", type="numpy"),
-    examples=[["./example.wav"]],
     title=title,
     description=description,
 )
 
 with demo:
-    gr.TabbedInterface([mic_translate, file_translate], ["Microphone", "Audio File"])
+    gr.TabbedInterface([mic_translate, file_translate], ["Microphone", "File"])
 
 demo.launch()