app.py

"""ML_task3.ipynb
Automatically generated by Colaboratory.
Original file is located at
    https://colab.research.google.com/drive/1DfK6fjkAd9RjVx3MUGfDtAOulvEenk0E?authuser=0#scrollTo=x0BqFNWue3V8
"""

import gradio as gr
import numpy as np
import torch
from datasets import load_dataset

from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline, WhisperProcessor, VitsModel, VitsTokenizer

device = "cuda:0" if torch.cuda.is_available() else "cpu"

# распознавание речи
asr_pipe = pipeline("automatic-speech-recognition", model="voidful/wav2vec2-xlsr-multilingual-56", device=device)

processor = WhisperProcessor.from_pretrained(
    "openai/whisper-small")

translator_en = pipeline("translation", model="Helsinki-NLP/opus-mt-mul-en")
translator_ru = pipeline("translation", model="Helsinki-NLP/opus-mt-en-ru")


model = VitsModel.from_pretrained("facebook/mms-tts-rus")
tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-rus")


def translator_mul_ru(text):

    translation  = translator_ru(translator_en(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):
    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[0]


title = "Speech-to-Speech Translation"
description = """
* Выбранная ASR модель - https://huggingface.co/voidful/wav2vec2-xlsr-multilingual-56
* Перевод текста на русский с помощью модели https://huggingface.co/Helsinki-NLP/opus-mt-mul-en 
* Синтез речи на русском языке с помощью модели https://huggingface.co/facebook/mms-tts-rus
"""

demo = gr.Blocks()

mic_translate = gr.Interface(
    fn=speech_to_speech_translation,
    inputs=gr.Audio(source="microphone", type="filepath"),
    outputs=gr.Audio(label="Generated Speech", type="numpy"),
    title=title,
    description=description,
)

file_translate = gr.Interface(
    fn=speech_to_speech_translation,
    inputs=gr.Audio(source="upload", type="filepath"),
    outputs=gr.Audio(label="Generated Speech", type="numpy"),
    title=title,
    description=description,
)

with demo:
    gr.TabbedInterface([mic_translate, file_translate], ["Microphone", "File"])

demo.launch()