import gradio as gr
import torch
import soundfile as sf
from qwen_tts import Qwen3TTSModel
from langdetect import detect
import os

# Load model - optimized for BF16 to save memory
device = "cuda" if torch.cuda.is_available() else "cpu"
model_id = "Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice"

print(f"Loading model to {device}...")
model = Qwen3TTSModel.from_pretrained(
    model_id,
    device_map=device,
    torch_dtype=torch.bfloat16 if device == "cuda" else torch.float32
)

def smart_tts(text, voice, instructions, auto_detect):
    try:
        # Smart Language Detection
        lang_map = {
            'zh': 'Chinese', 'en': 'English', 'jp': 'Japanese', 
            'ko': 'Korean', 'de': 'German', 'fr': 'French', 
            'ru': 'Russian', 'pt': 'Portuguese', 'es': 'Spanish', 'it': 'Italian'
        }
        
        detected_lang = "English" # Default
        if auto_detect:
            try:
                raw_lang = detect(text).split('-')[0]
                detected_lang = lang_map.get(raw_lang, "English")
            except:
                pass

        # Generate Audio
        # The CustomVoice model uses instructions for style/emotion
        wavs, sr = model.generate_custom_voice(
            language=detected_lang,
            speaker=voice,
            instruct=instructions,
            text=text
        )
        
        output_path = "output.wav"
        sf.write(output_path, wavs[0], sr)
        return output_path, f"Detected Language: {detected_lang}"
        
    except Exception as e:
        return None, str(e)

# UI Layout
with gr.Blocks(theme=gr.themes.Soft()) as demo:
    gr.Markdown(f"# 🗣️ Qwen3-TTS Smart Studio")
    gr.Markdown("Experience natural speech with style control using Qwen3-TTS-12Hz.")
    
    with gr.Row():
        with gr.Column():
            input_text = gr.Textbox(
                label="Input Text", 
                placeholder="Type something here...",
                lines=4
            )
            
            with gr.Row():
                voice_select = gr.Dropdown(
                    choices=["Vivian", "Ryan", "Bella", "Daisy", "George"], 
                    value="Vivian", 
                    label="Speaker"
                )
                auto_lang = gr.Checkbox(label="Auto-detect Language", value=True)
            
            style_instruct = gr.Textbox(
                label="Style Instruction (e.g., 'Speak with a happy tone')", 
                placeholder="Angry, Sad, Excited, Whisper...",
                value="Speak naturally"
            )
            
            generate_btn = gr.Button("Generate Speech", variant="primary")
            
        with gr.Column():
            audio_output = gr.Audio(label="Generated Audio", type="filepath")
            status_info = gr.Label(label="System Status")

    generate_btn.click(
        fn=smart_tts,
        inputs=[input_text, voice_select, style_instruct, auto_lang],
        outputs=[audio_output, status_info]
    )

if __name__ == "__main__":
    demo.launch()