STST / app.py
pragsGit's picture
Update app.py
90f8151 verified
import gradio as gr
import numpy as np
import torch
from datasets import load_dataset
from transformers import pipeline
from transformers import VitsModel, VitsTokenizer
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)
model = VitsModel.from_pretrained("facebook/mms-tts-spa")
processor = VitsTokenizer.from_pretrained("facebook/mms-tts-spa")
def translate(audio):
outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"language": "es","task": "transcribe"})
return outputs["text"]
def synthesise(text):
inputs = processor(text=text, return_tensors="pt")
with torch.no_grad():
speech = model(inputs["input_ids"].to(device))
return speech.audio[0]
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
title = "Cascaded STST"
description = """
Demo for cascaded speech-to-speech translation (STST), mapping from source speech in any language to target speech in English. Demo uses OpenAI's [Whisper Base](https://huggingface.co/openai/whisper-base) model for speech translation, and Microsoft's
[SpeechT5 TTS](https://huggingface.co/microsoft/speecht5_tts) model for text-to-speech:
![Cascaded STST](https://huggingface.co/datasets/huggingface-course/audio-course-images/resolve/main/s2st_cascaded.png "Diagram of cascaded speech to speech translation")
"""
demo = gr.Blocks()
mic_translate = gr.Interface(
fn=speech_to_speech_translation,
inputs=gr.Audio(sources="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(sources="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()