import gradio as gr
import requests
import random
import urllib.parse
import tempfile
import os

NSFW_URL_TEMPLATE = os.getenv("NSFW_API_URL_TEMPLATE")
TTS_URL_TEMPLATE = os.getenv("TTS_API_URL_TEMPLATE")

VOICES = [
    "alloy", "echo", "fable", "onyx", "nova", "shimmer",
    "coral", "verse", "ballad", "ash", "sage", "amuch", "dan"
]

def check_nsfw(prompt: str) -> bool:
    try:
        encoded_prompt = urllib.parse.quote(prompt)
        url = NSFW_URL_TEMPLATE.format(prompt=encoded_prompt)
        response = requests.get(url, timeout=20)
        result = response.text.strip().upper()
        return result == "YES"
    except:
        return True  # assume NSFW if check fails

def generate_audio(prompt: str, voice: str, emotion: str, seed: int) -> bytes:
    try:
        encoded_prompt = urllib.parse.quote(prompt)
        encoded_emotion = urllib.parse.quote(emotion)
        url = TTS_URL_TEMPLATE.format(
            prompt=encoded_prompt, emotion=encoded_emotion, voice=voice, seed=seed
        )
        response = requests.get(url, timeout=60)
        response.raise_for_status()

        if 'audio' not in response.headers.get('content-type', '').lower():
            raise gr.Error("API response is not audio.")
        return response.content

    except Exception as e:
        print("Error:", e)
        raise gr.Error("Error generating audio.")

def text_to_speech_app(prompt, voice, emotion, use_random_seed, specific_seed):
    if not prompt: raise gr.Error("Prompt cannot be empty.")
    if not voice: raise gr.Error("Voice selection required.")
    if not emotion: emotion = "neutral"

    seed = random.randint(0, 2**32 - 1) if use_random_seed else int(specific_seed)
    is_nsfw = False  # Set to check_nsfw(prompt) if needed

    if is_nsfw:
        return None, "⚠️ Inappropriate prompt detected."

    audio_bytes = generate_audio(prompt, voice, emotion, seed)
    with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp:
        tmp.write(audio_bytes)
        return tmp.name, f"✅ Audio ready | Voice: {voice} | Emotion: {emotion} | Seed: {seed}"

def toggle_seed_input(use_random_seed):
    return gr.update(visible=not use_random_seed, value=12345)

with gr.Blocks() as app:
    gr.Markdown("## 🎙️ DeepCAL Oracle Voice Engine")
    gr.Markdown("Enter text, select a voice and emotion, then generate oracle-grade audio.")

    with gr.Row():
        with gr.Column(scale=2):
            prompt_input = gr.Textbox(label="Prompt")
            emotion_input = gr.Textbox(label="Emotion", placeholder="e.g., sarcastic, serious")
            voice_dropdown = gr.Dropdown(label="Voice", choices=VOICES, value="alloy")
        with gr.Column(scale=1):
            random_seed_checkbox = gr.Checkbox(label="Use Random Seed", value=True)
            seed_input = gr.Number(label="Specific Seed", visible=False, value=12345)

    submit_button = gr.Button("🎧 Generate Audio")

    with gr.Row():
        audio_output = gr.Audio(label="Audio Output", type="filepath")
        status_output = gr.Textbox(label="Status")

    random_seed_checkbox.change(toggle_seed_input, [random_seed_checkbox], [seed_input])

    submit_button.click(
        fn=text_to_speech_app,
        inputs=[prompt_input, voice_dropdown, emotion_input, random_seed_checkbox, seed_input],
        outputs=[audio_output, status_output],
        concurrency_limit=30
    )

    gr.Examples(
        examples=[
            ["Welcome to DeepCAL Oracle.", "ballad", "wise, cosmic", True, 12345],
            ["Headshot logistics incoming.", "shimmer", "sarcastic and aggressive", True, 5555],
        ],
        inputs=[prompt_input, voice_dropdown, emotion_input, random_seed_checkbox, seed_input],
        outputs=[audio_output, status_output],
        fn=text_to_speech_app,
        cache_examples=False
    )

if __name__ == "__main__":
    if NSFW_URL_TEMPLATE and TTS_URL_TEMPLATE:
        app.launch()
    else:
        print("🚫 Missing required secrets. Please set NSFW_API_URL_TEMPLATE and TTS_API_URL_TEMPLATE.")