File size: 2,865 Bytes
d41c1a4
 
 
 
 
 
 
 
e2334e5
d41c1a4
 
 
e2334e5
8e253e0
d41c1a4
 
 
 
 
 
 
a824dd6
d41c1a4
 
 
665125d
d41c1a4
 
 
 
 
 
b28d750
 
d41c1a4
 
 
665125d
d41c1a4
 
 
 
 
 
 
 
 
 
 
 
96203f7
 
4fbf54b
d41c1a4
 
 
 
 
 
b27d944
d41c1a4
 
 
 
 
 
 
b27d944
d41c1a4
b27d944
d41c1a4
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
# -*- coding: utf-8 -*-
"""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"

#eng audio to text transformation
asr_pipe = pipeline("automatic-speech-recognition", model="asapp/sew-d-tiny-100k-ft-ls100h", device=device)

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

#rus text to rus speech transformation
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 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]

title = "Cascaded STST"
description = """
В Демо 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 facebook/wmt19-en-ru(https://huggingface.co/facebook/wmt19-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=[["./example.wav"]],
    title=title,
    description=description,
)

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

demo.launch()