app.py

import gradio as gr
import numpy as np
import torch
from transformers import pipeline, VitsModel, VitsTokenizer

device = "cuda:0" if torch.cuda.is_available() else "cpu"

target_dtype = np.int16
max_range = np.iinfo(target_dtype).max

# load speech translation checkpoint
ASR_MODEL_NAME = 'openai/whisper-base'
asr_pipe = pipeline("automatic-speech-recognition", model=ASR_MODEL_NAME, device=device)


# load text-to-speech checkpoint
model = VitsModel.from_pretrained("Matthijs/mms-tts-deu")
tokenizer = VitsTokenizer.from_pretrained("Matthijs/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.audio[0]
    return speech.cpu()
    
def speech_to_speech_translation(audio):
    translated_text = translate(audio)
    synthesised_speech = synthesise(translated_text)
    synthesised_speech = (synthesised_speech.numpy() * max_range).astype(np.int16)
    return 16000, synthesised_speech
    
    
title = "Cascaded STST - Any language to German speech"
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 Microsoft's
[MMS TTS](https://huggingface.co/Matthijs/mms-tts-deu) 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.queue(concurrency_count=2,max_size=10)
demo.launch()