|
import torch |
|
import gradio as gr |
|
import numpy as np |
|
|
|
from transformers import ( |
|
VitsModel, |
|
VitsTokenizer, |
|
pipeline, |
|
) |
|
|
|
|
|
device = "cuda:0" if torch.cuda.is_available() else "cpu" |
|
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32 |
|
print(f"Using {device} with fp {torch_dtype}") |
|
|
|
|
|
asr_pipe = pipeline( |
|
"automatic-speech-recognition", |
|
model="openai/whisper-medium", |
|
device=device, |
|
torch_dtype=torch_dtype, |
|
) |
|
|
|
|
|
|
|
model = VitsModel.from_pretrained("facebook/mms-tts-zlm") |
|
tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-zlm") |
|
|
|
|
|
def synthesise(text): |
|
inputs = tokenizer(text=text, return_tensors="pt") |
|
input_ids = inputs["input_ids"] |
|
|
|
with torch.no_grad(): |
|
outputs = model(input_ids) |
|
|
|
speech = outputs["waveform"] |
|
return speech |
|
|
|
|
|
def translate(audio): |
|
outputs = asr_pipe( |
|
audio, |
|
max_new_tokens=256, |
|
generate_kwargs={"task": "transcribe", "language": "ms"}, |
|
) |
|
return outputs["text"] |
|
|
|
|
|
def speech_to_speech_translation(audio): |
|
translated_text = translate(audio) |
|
synthesised_speech = synthesise(translated_text) |
|
synthesised_speech = (synthesised_speech.numpy() * 32767).astype(np.int16) |
|
return 16000, synthesised_speech.T |
|
|
|
|
|
title = "Cascaded STST" |
|
description = """ |
|
Demo for cascaded speech-to-speech translation (STST), mapping from source speech in any language to target speech in **Malay**. Demo uses OpenAI's [Whisper Base](https://huggingface.co/openai/whisper-base) model for speech translation, and Facebooks's |
|
[MMS-TTS-ZLM](https://huggingface.co/facebook/mms-tts-zlm) model for text-to-speech: |
|
|
|
 |
|
""" |
|
|
|
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="./examples", |
|
title=title, |
|
description=description, |
|
live=True, |
|
) |
|
|
|
with demo: |
|
gr.TabbedInterface([mic_translate, file_translate], ["Microphone", "Audio File"]) |
|
|
|
demo.launch() |
|
|
