|
|
|
"""app.ipynb |
|
|
|
Automatically generated by Colaboratory. |
|
|
|
Original file is located at |
|
https://colab.research.google.com/drive/16MxXQeF3O0htL9eQ61aa6ZxnApGg9TKN |
|
""" |
|
|
|
import gradio as gr |
|
import numpy as np |
|
import torch |
|
|
|
from transformers import pipeline, VitsModel, VitsTokenizer, FSMTForConditionalGeneration, FSMTTokenizer |
|
|
|
device = "cuda:0" if torch.cuda.is_available() else "cpu" |
|
|
|
|
|
asr_pipe = pipeline("automatic-speech-recognition", model="asapp/sew-d-tiny-100k-ft-ls100h", device=device) |
|
|
|
|
|
translation_pipe = pipeline("translation", model="Helsinki-NLP/opus-mt-en-ru") |
|
|
|
|
|
vits_model = VitsModel.from_pretrained("facebook/mms-tts-rus") |
|
vits_tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-rus") |
|
|
|
def transform_audio_to_speech_en(audio): |
|
outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"}) |
|
return outputs["text"] |
|
|
|
def translator(text): |
|
translated_text = translation_pipe(text) |
|
return translated_text[0]['translation_text'] |
|
|
|
def transform_audio_to_speech_ru(translated_text): |
|
translated_text = translator(translated_text) |
|
inputs = vits_tokenizer(translated_text, return_tensors="pt") |
|
with torch.no_grad(): |
|
speech = vits_model(**inputs).waveform |
|
return speech.cpu() |
|
|
|
def speech_to_speech_translation(audio): |
|
translated_text = transform_audio_to_speech_en(audio) |
|
synthesised_speech = transform_audio_to_speech_ru(translated_text) |
|
synthesised_speech = (synthesised_speech.numpy() * 32767).astype(np.int16) |
|
return 16000, synthesised_speech[0] |
|
|
|
title = "Cascaded STST" |
|
description = """ |
|
В приложении используется модель SEW-D-tiny(https://huggingface.co/asapp/sew-d-tiny-100k-ft-ls100h), |
|
распознающая английскую речь и преобразующая ее в строку. Затем с помощью модели Helsinki-NLP/opus-mt-en-ru(https://huggingface.co/Helsinki-NLP/opus-mt-en-ru) текст |
|
переводится на русский язык и преобразуется в русскую речь с помощью модели facebook/mms-tts-rus(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"), |
|
examples=[["./personal_example.wav"]], |
|
title=title, |
|
description=description, |
|
) |
|
|
|
with demo: |
|
gr.TabbedInterface([mic_translate, file_translate], ["Microphone", "Audio File"]) |
|
|
|
demo.launch() |
