app.py

import gradio as gr
from transformers import pipeline
from gtts import gTTS
import tempfile

# Load ASR & Translation models
asr_model = pipeline("automatic-speech-recognition", model="openai/whisper-medium", device=0)
translator = pipeline("translation", model="ai4bharat/indictrans2-en-indic", device=0)

# Map language names to codes
lang_code_map = {
    "Hindi": "hi", "Kannada": "kn", "Tamil": "ta", "Telugu": "te", "Bengali": "bn", "Marathi": "mr",
    "Gujarati": "gu", "Punjabi": "pa", "Malayalam": "ml", "Assamese": "as", "Odia": "or", "Urdu": "ur",
    "Sanskrit": "sa", "Nepali": "ne", "Manipuri": "mni", "English": "en"
}

languages = list(lang_code_map.keys())

def process_speech(audio, speaker_lang, listener_lang):
    # Step 1: Speech → Text
    transcript = asr_model(audio)["text"]

    # Step 2: Translate
    try:
        result = translator(transcript, src_lang=speaker_lang.lower(), tgt_lang=listener_lang.lower())
        translated_text = result[0]["translation_text"]
    except:
        translated_text = "[Translation failed]"

    # Step 3: Text → Speech
    tts_lang = lang_code_map.get(listener_lang, "en")
    tts = gTTS(translated_text, lang=tts_lang)
    with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as f:
        tts.save(f.name)
        return transcript, translated_text, f.name

# Gradio UI
with gr.Blocks(title="Indian Language Speech Translator") as demo:
    gr.Markdown("## 🗣️ Real-Time Indian Speech Translator")
    gr.Markdown("Choose two languages and take turns speaking. The app will translate and speak for you in real-time.")

    with gr.Row():
        speaker_lang = gr.Dropdown(choices=languages, value="Hindi", label="🎙️ Speaker's Language")
        listener_lang = gr.Dropdown(choices=languages, value="Kannada", label="👂 Listener's Language")

    audio_input = gr.Audio(source="microphone", type="filepath", label="🎤 Speak here")

    with gr.Row():
        transcript_box = gr.Textbox(label="📝 Recognized Speech")
        translation_box = gr.Textbox(label="🌐 Translated Text")

    audio_output = gr.Audio(label="🔊 Translated Speech")

    btn = gr.Button("Translate & Speak")

    btn.click(
        fn=process_speech,
        inputs=[audio_input, speaker_lang, listener_lang],
        outputs=[transcript_box, translation_box, audio_output]
    )

demo.launch()