app.py

import gradio as gr
import tempfile
import torch

# =========================
# ASR (Faster Whisper - CPU)
# =========================
from faster_whisper import WhisperModel

asr_model = WhisperModel(
    "base", 
    device="cpu", 
    compute_type="int8"
)

def transcribe(audio):
    if audio is None:
        return ""
    
    segments, _ = asr_model.transcribe(audio)
    text = " ".join([seg.text for seg in segments])
    return text.strip()


# =========================
# LLM (Qwen 0.5B - CPU)
# =========================
from transformers import AutoTokenizer, AutoModelForCausalLM

MODEL_NAME = "Qwen/Qwen2.5-0.5B-Instruct"

tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)

model = AutoModelForCausalLM.from_pretrained(
    MODEL_NAME,
    dtype=torch.float32,
    low_cpu_mem_usage=True
)

def generate_response(text):
    if not text:
        return "Say something..."

    prompt = f"User: {text}\nAssistant:"
    
    inputs = tokenizer(prompt, return_tensors="pt")

    with torch.no_grad():
        output = model.generate(
            **inputs,
            max_new_tokens=120,
            do_sample=True,
            temperature=0.7,
            top_p=0.9
        )

    response = tokenizer.decode(output[0], skip_special_tokens=True)
    return response.split("Assistant:")[-1].strip()


# =========================
# TTS (CPU - pyttsx3)
# =========================
import pyttsx3

engine = pyttsx3.init()
engine.setProperty("rate", 170)

def text_to_speech(text):
    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as f:
        path = f.name
    
    engine.save_to_file(text, path)
    engine.runAndWait()
    return path


# =========================
# FULL PIPELINE
# =========================
def full_pipeline(audio):
    text = transcribe(audio)
    response = generate_response(text)
    audio_out = text_to_speech(response)

    return text, response, audio_out


# =========================
# GRADIO UI
# =========================
with gr.Blocks() as demo:
    gr.Markdown("# 🎙️ CPU Voice Agent (ASR + LLM + TTS)")

    audio_input = gr.Audio(
        sources=["microphone"], 
        type="filepath",
        label="Speak"
    )

    btn = gr.Button("Run")

    text_out = gr.Textbox(label="Transcription")
    response_out = gr.Textbox(label="LLM Response")
    audio_out = gr.Audio(label="Response Audio")

    btn.click(
        fn=full_pipeline,
        inputs=audio_input,
        outputs=[text_out, response_out, audio_out]
    )

demo.launch()