app.py

import gradio as gr
import torchaudio
import torch


def speech_to_text(audio_file):
    audio_input, _ = torchaudio.load(audio_file.name)
    s2t_model = torch.jit.load("unity_on_device_s2t.ptl")
    with torch.no_grad():
        text = s2t_model(audio_input, tgt_lang=TGT_LANG)
    return text

def speech_to_speech_translation(audio_file):
    audio_input, _ = torchaudio.load(audio_file.name)
    s2st_model = torch.jit.load("unity_on_device.ptl")
    with torch.no_grad():
        text, units, waveform = s2st_model(audio_input, tgt_lang=TGT_LANG)
    output_file = "/tmp/result.wav"
    torchaudio.save(output_file, waveform.unsqueeze(0), sample_rate=16000)
    return text, output_file

# Gradio interfaces
iface_s2t = gr.Interface(
    fn=speech_to_text,
    inputs=gr.Audio(type="file", label="Upload Audio for Speech to Text"),
    outputs="text",
    title="Speech to Text"
)

iface_s2st = gr.Interface(
    fn=speech_to_speech_translation,
    inputs=gr.Audio(type="file", label="Upload Audio for Speech to Speech Translation"),
    outputs=["text", "audio"],
    title="Speech to Speech Translation"
)

# Combine into an accordion interface
accordion = gr.Accordion(
    iface_s2t,
    iface_s2st,
    labels=["Speech to Text", "Speech to Speech Translation"]
)

# Launch the application
accordion.launch()