app.py

import gradio as gr
import os
import whisper
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
from gtts import gTTS
from tempfile import NamedTemporaryFile

# Define translation function
def translate_audio(input_file, target_language):
    # Save uploaded audio file to a temporary file
    with NamedTemporaryFile(suffix=".mp3") as temp_audio:
        temp_audio.write(input_file.read())
        temp_audio.seek(0)
    
    # Auto to text (STT)
    model = whisper.load_model("base")
    audio = whisper.load_audio(temp_audio.name)
    audio = whisper.pad_or_trim(audio)
    mel = whisper.log_mel_spectrogram(audio).to(model.device)
    _, probs = model.detect_language(mel)
    options = whisper.DecodingOptions()
    result = whisper.decode(model, mel, options)
    text = result.text
    lang = max(probs, key=probs.get)

    # Translate
    tokenizer = AutoTokenizer.from_pretrained("alirezamsh/small100")
    model = AutoModelForSeq2SeqLM.from_pretrained("alirezamsh/small100")
    tokenizer.src_lang = target_language
    encoded_bg = tokenizer(text, return_tensors="pt")
    generated_tokens = model.generate(**encoded_bg)
    translated_text = tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)[0]

    # Text-to-audio (TTS)
    tts = gTTS(text=translated_text, lang=target_language)
    output_file = NamedTemporaryFile(suffix=".mp3", delete=False)
    output_file.close()
    tts.save(output_file.name)
    return output_file.name

# Define Gradio interface
inputs = [
 gr.File(label="Upload Audio File"),
 gr.Dropdown(choices=['en', 'es', 'fr', 'de', 'ru'], label="Target Language")
]

outputs = [
 gr.File(label="Translated Audio")
]

title = "Audio Translation"
description = "Upload an audio file, translate the speech to a target language, and download the translated audio."

gr.Interface(fn=translate_audio, inputs=inputs, outputs=outputs, title=title, description=description).launch()