import os
import gradio as gr
import whisper
from groq import Groq
from gtts import gTTS
import tempfile

# Load Groq API key from Hugging Face secrets
GROQ_API_KEY = "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"

client = Groq(api_key=GROQ_API_KEY)

# Load Whisper model
whisper_model = whisper.load_model("base")

def transcribe_and_chat(audio_file):
    if audio_file is None:
        return "No audio received", "Error: Please record again.", None

    try:
        # 1. Transcribe
        result = whisper_model.transcribe(audio_file)
        user_text = result.get("text", "").strip()
        if not user_text:
            return "Could not transcribe", "Error: Empty transcription", None

        # 2. Query Groq
        chat_completion = client.chat.completions.create(
            messages=[{"role": "user", "content": user_text}],
            model="llama-3.1-8b-instant"  # ✅ stable model
        )
        bot_text = chat_completion.choices[0].message.content

        # 3. TTS
        tts = gTTS(bot_text)
        with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp:
            tts.save(tmp.name)
            reply_audio = tmp.name

        return user_text, bot_text, reply_audio

    except Exception as e:
        return "Transcription error", str(e), None

# Gradio UI
with gr.Blocks() as demo:
    gr.Markdown("## 🎙️ Real-time Voice-to-Voice Chatbot (Whisper + Groq + gTTS)")

    audio_in = gr.Audio(sources=["microphone"], type="filepath", label="🎤 Speak here")
    user_text_out = gr.Textbox(label="📝 Transcription (You said)")
    bot_text_out = gr.Textbox(label="🤖 LLM Reply")
    audio_out = gr.Audio(label="🔊 Bot's Voice", type="filepath")

    btn = gr.Button("▶️ Process")
    btn.click(
        fn=transcribe_and_chat,
        inputs=[audio_in],
        outputs=[user_text_out, bot_text_out, audio_out]
    )

demo.launch()