app.py

import gradio as gr
import numpy as np
import torch

from transformers import VitsModel, AutoTokenizer, pipeline


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

# load text-to-speech checkpoint
model = VitsModel.from_pretrained("facebook/mms-tts-fra")
tokenizer = AutoTokenizer.from_pretrained("facebook/mms-tts-fra")


def translate(audio):
    outputs = asr_pipe(
        audio,
        max_new_tokens=512,
        generate_kwargs={"task": "transcribe", "language": "fr"}
    )
    return outputs["text"]


def synthesise(text):
    inputs = tokenizer(text=text, return_tensors="pt")
    with torch.no_grad():
        speech = model(**inputs).waveform
    speech = speech[0]  # remove batch dimension
    return speech


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 French. Demo uses OpenAI's [Whisper Base](https://huggingface.co/openai/whisper-base) model for speech translation, and Facebook's
[MMS TTS French](https://huggingface.co/facebook/mms-tts-fra) 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()