app.py

import gradio as gr
from huggingface_hub import InferenceClient
import torch
from TTS.api import TTS
import os
import subprocess

# Force CPU usage
device = "cpu"

# Load TTS Model (Running on CPU)
tts_model = TTS("tts_models/en/ljspeech/tacotron2-DDC", gpu=False)  # ✅ Ensures CPU-only execution

# Hugging Face LLM Client (DeepSeek R1 7B)
client = InferenceClient("deepseek-ai/deepseek-r1-7b")

# RVC Model Paths
RVC_MODEL_PATH = "zeldabotw.pth"
RVC_INDEX_PATH = "zeldabotw.index"

# Function to call RVC for voice conversion (CPU Mode)
def convert_voice(input_wav, output_wav):
    """Converts input TTS audio to ZeldaBotW voice using RVC (CPU Mode)."""
    if not os.path.exists(RVC_MODEL_PATH) or not os.path.exists(RVC_INDEX_PATH):
        raise FileNotFoundError("RVC model files not found! Ensure zeldabotw.pth and zeldabotw.index are in the same directory.")

    command = f"python infer_rvc.py --input {input_wav} --output {output_wav} --model {RVC_MODEL_PATH} --index {RVC_INDEX_PATH} --pitch_shift 0 --device cpu"
    
    process = subprocess.run(command, shell=True, capture_output=True, text=True)
    if process.returncode != 0:
        print("RVC conversion failed:", process.stderr)
        return None
    return output_wav

# Chatbot Response + TTS + RVC
def respond(message, history, system_message, max_tokens, temperature, top_p):
    messages = [{"role": "system", "content": system_message}]
    
    for val in history:
        if val[0]: messages.append({"role": "user", "content": val[0]})
        if val[1]: messages.append({"role": "assistant", "content": val[1]})
    
    messages.append({"role": "user", "content": message})
    
    response = ""

    # Get LLM Response
    for message in client.chat_completion(messages, max_tokens=max_tokens, stream=False, temperature=temperature, top_p=top_p):
        response += message.choices[0].message.content

    # Generate Speech from Text (CPU Mode)
    tts_audio_path = "tts_output.wav"
    tts_model.tts_to_file(text=response, file_path=tts_audio_path)

    # Convert TTS output to ZeldaBotW voice (CPU Mode)
    rvc_audio_path = "rvc_output.wav"
    rvc_converted_path = convert_voice(tts_audio_path, rvc_audio_path)

    return response, tts_audio_path, rvc_converted_path  # ✅ Now correctly returns all outputs

# Gradio UI
with gr.Blocks() as demo:
    gr.Markdown("## DeepSeek R1 7B Chatbot with ZeldaBotW Voice (CPU Mode)")
    
    chatbot = gr.Chatbot(type="messages")  # ✅ Fix deprecated type warning
    msg = gr.Textbox(label="User Input")
    
    system_msg = gr.Textbox(value="You are a friendly Chatbot.", label="System Message")
    max_tokens = gr.Slider(1, 2048, value=512, step=1, label="Max Tokens")
    temperature = gr.Slider(0.1, 4.0, value=0.7, step=0.1, label="Temperature")
    top_p = gr.Slider(0.1, 1.0, value=0.95, step=0.05, label="Top-p (Nucleus Sampling)")
    
    tts_audio = gr.Audio(type="filepath", label="TTS Output")
    rvc_audio = gr.Audio(type="filepath", label="RVC ZeldaBotW Voice")

    def chat_fn(message, history):
        return respond(message, history, system_msg.value, max_tokens.value, temperature.value, top_p.value)

    msg.submit(chat_fn, inputs=[msg, chatbot], outputs=[chatbot, tts_audio, rvc_audio])

demo.launch()