|
import gradio as gr |
|
import numpy as np |
|
import torch |
|
from datasets import load_dataset |
|
|
|
from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline, VitsModel, VitsTokenizer |
|
|
|
|
|
device = "cuda:0" if torch.cuda.is_available() else "cpu" |
|
|
|
|
|
asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device) |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
model = VitsModel.from_pretrained('facebook/mms-tts-ron') |
|
tokenizer = VitsTokenizer.from_pretrained('facebook/mms-tts-ron') |
|
|
|
embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation") |
|
speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0) |
|
|
|
|
|
def translate(audio): |
|
outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "ro"}) |
|
return outputs["text"] |
|
|
|
|
|
def synthesise(text): |
|
inputs = tokenizer(text=text, return_tensors="pt") |
|
|
|
with torch.no_grad(): |
|
tts_output = model(inputs['input_ids'].to(device)) |
|
|
|
speech = tts_output['waveform'] |
|
return speech.cpu() |
|
|
|
|
|
def speech_to_speech_translation(audio): |
|
translated_text = translate(audio) |
|
print(translated_text) |
|
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 Romanian. Demo uses OpenAI's [Whisper Base](https://huggingface.co/openai/whisper-base) model for speech translation, and Facebook's |
|
[MMS-TTS-RON](https://huggingface.co/facebook/mms-tts-ron) 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(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() |
|
|