app.py

"""
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
from transformers import VitsModel, VitsTokenizer, FSMTForConditionalGeneration, FSMTTokenizer, Wav2Vec2ForCTC, Wav2Vec2Processor, MarianMTModel, MarianTokenizer, T5ForConditionalGeneration, T5Tokenizer

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

# Transform audio to en text
asr_pipe = pipeline("automatic-speech-recognition", model="asapp/sew-d-mid-400k-ft-ls100h", device=device)

# Translate en to rus text
translation_en_to_rus = pipeline("translation", model="Helsinki-NLP/opus-mt-en-ru")

# Create speech from rus text
#vits_model = VitsModel.from_pretrained("Helsinki-NLP/opus-mt-ru-en")
#vits_tokenizer = VitsTokenizer.from_pretrained("Helsinki-NLP/opus-mt-ru-en")

model = MarianMTModel.from_pretrained("Helsinki-NLP/opus-mt-ru-en")
tokenizer = MarianTokenizer.from_pretrained("Helsinki-NLP/opus-mt-ru-en")

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 = translator_model.generate(**translator_tokenizer(text, return_tensors="pt", padding=True))
    translated_text = translator_tokenizer.decode(translated[0], skip_special_tokens=True)
    return translated_text

def synthesise(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 = synthesise(translated_text)
    synthesised_speech = (synthesised_speech.numpy() * 32767).astype(np.int16)
    return 16000, synthesised_speech[0]


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")
)

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"]]
)

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

demo.launch()