File size: 2,738 Bytes
aa2ad2d b0361e0 aa2ad2d b0361e0 aa2ad2d b0361e0 aa2ad2d b0361e0 aa2ad2d c705c41 aa2ad2d c705c41 aa2ad2d b0361e0 aa2ad2d b0361e0 aa2ad2d 58e77db 92e4310 58e77db aa2ad2d 58e77db aa2ad2d |
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 79 80 |
import gradio as gr
import numpy as np
import torch
from datasets import load_dataset
from transformers import VitsModel, VitsTokenizer, pipeline
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)
# loading the deutsch multilingual checkpoint
model = VitsModel.from_pretrained("facebook/mms-tts-deu")
tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-deu")
def translate(audio):
outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe" , "language": "de"})
return outputs["text"]
def synthesise(text):
inputs = tokenizer(text, return_tensors="pt")
input_ids = inputs["input_ids"]
with torch.no_grad():
outputs = model(input_ids)
speech = outputs["waveform"]
return speech
# converting the output audio array to int16,which is expected by gradio
target_dtype = np.int16
max_range = np.iinfo(target_dtype).max
def speech_to_speech_translation(audio):
translated_text = translate(audio)
synthesised_speech = synthesise(translated_text)
# converting for gradio
synthesised_speech = (synthesised_speech.squeeze().numpy() * max_range).astype(np.int16)
return 16000, synthesised_speech
title = "Cascaded Speech To Speech Translation in German"
description = """
Demo for cascaded speech-to-speech translation (STST), mapping from source speech in any language to target speech in German. Demo uses OpenAI's [Whisper Base](https://huggingface.co/openai/whisper-base) model for speech translation, and Meta's [Massively Multilingual Speech German](https://huggingface.co/facebook/mms-tts-deu) model for text-to-speech.
The below diagram shows how the cascaded speech to speech translation works.
"""
demo = gr.Blocks()
mic_translate = gr.Interface(
fn=speech_to_speech_translation,
inputs=gr.Audio(sources="microphone",label= "Audio", 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(sources="upload", label="Audio file", type="filepath"),
outputs=gr.Audio(label="Generated Speech", type="numpy"),
examples=[["./example.wav"]],
title=title,
description=description,
cache_examples=True,
allow_flagging="never",
)
with demo:
gr.TabbedInterface([mic_translate,file_translate], ["Microphone","Audio File"])
demo.launch()
|