app.py

#!/usr/bin/env python3
"""DramaBox — Gradio demo (warm server).

Loads the warm TTSServer once, then handles requests at ~2.5 s each. All
generated audio is invisibly watermarked with Resemble Perth before being
returned to the user.
"""
import logging
import os
import sys
import tempfile
import time

import gradio as gr
from fastapi.responses import HTMLResponse
from fastapi.staticfiles import StaticFiles
from gradio import Server
from gradio.data_classes import FileData
import spaces


# Local src import.
sys.path.insert(0, os.path.join(os.path.dirname(os.path.abspath(__file__)), "src"))
from inference_server import TTSServer  # noqa: E402
from model_downloader import get_all_paths  # noqa: E402


logging.basicConfig(level=logging.INFO, format="%(asctime)s %(levelname)s %(message)s")
logging.info("Fetching DramaBox checkpoints from HuggingFace (cached after first run)...")
PATHS = get_all_paths()

# Module-level warm load (same pattern as IndexTTS-2-Demo on ZeroGPU). The
# `spaces` package patches torch so that .to("cuda") at import time pins the
# weights into ZeroGPU's shared memory; each @spaces.GPU call then maps them
# onto the actual GPU instantly. First user request is ~2.5 s instead of ~30 s.
logging.info("Loading DramaBox warm server (Gemma + DiT + VAE + Decoder)...")
tts = TTSServer(
    checkpoint=PATHS["transformer"],
    full_checkpoint=PATHS["audio_components"],
    gemma_root=PATHS["gemma_root"],
    device="cuda",
    dtype=os.environ.get("LTX_DTYPE", "bf16"),
    compile_model=False,                  # torch.compile breaks under ZeroGPU's brief GPU windows
    bnb_4bit=True,                        # unsloth Gemma is pre-quantized
)
logging.info("TTSServer ready.")


# ── Example prompts shipped with a matching voice reference ──────────────────
# Files live under assets/voices/ so users can click a row and generate
# without uploading anything.
_VOICES_DIR = os.path.join(os.path.dirname(os.path.abspath(__file__)), "assets", "voices")

EXAMPLES: list[tuple[str, str, str]] = [
    (
        "Villain monologue",
        os.path.join(_VOICES_DIR, "male_harvey_keitel.mp3"),
        'A shadowy villain speaks with cold menace, "You have entered my domain, mortal." '
        'He chuckles darkly, "Such arrogance will be your undoing." '
        'His voice rises with fury, "Kneel, or be destroyed where you stand!"',
    ),
    (
        "Talk-show host wheeze-laugh",
        os.path.join(_VOICES_DIR, "male_conan.mp3"),
        'A talk show host gasps with shock, "No! You did NOT just say that!" '
        'He bursts into uncontrollable laughter, "Hahaha! Oh my god, oh my god!" '
        'He wheezes, "I cannot, I literally cannot breathe right now!"',
    ),
    (
        "Tender goodnight whisper",
        os.path.join(_VOICES_DIR, "female_shadowheart.wav"),
        'A woman speaks tenderly, "It has been a long day, my love." '
        'She whispers, "Close your eyes. I am right here." '
        'She hums quietly, "Mmmm-mmm. Sleep now."',
    ),
    (
        "Old-school radio anchor",
        os.path.join(_VOICES_DIR, "male_old_movie.wav"),
        'A radio host clears his throat, "Excuse me, pardon that." '
        'He settles into a warm, professional tone, "Good evening everyone, '
        'and welcome back to the show. We have got a wonderful lineup tonight."',
    ),
    (
        "Catgirl uncontrollable giggling",
        os.path.join(_VOICES_DIR, "female_american.wav"),
        'A playful girl already mid-giggle, "Hehehe, oh my gosh you should see your face!" '
        'She gasps for air between giggles, "Oh my, hehe, oh my, I cannot stop!" '
        'She tries to compose herself, "Ahhhhh okay okay okay, I will stop, I promise."',
    ),
    (
        "Hero stammering courage",
        os.path.join(_VOICES_DIR, "male_arnie.mp3"),
        'A young warrior speaks with a trembling voice, "I... I do not know if I can do this." '
        'He takes a shaky breath, "But someone has to try." '
        'His voice steadies with growing fire, "No more running. I WILL fight!"',
    ),
    (
        "Exhausted dad, fraying patience",
        os.path.join(_VOICES_DIR, "male_petergriffin.wav"),
        'An exhausted father speaks with fraying patience, "Sweetie, daddy is asking very nicely." '
        'He sighs deeply, "Ohhhh my goodness." '
        'He puts on an overly cheerful voice, "Hey buddy! Look at the shiny thing!" '
        'Then he laughs helplessly, "Hahaha, I am losing my mind."',
    ),
    (
        "Smug-confident announcer",
        os.path.join(_VOICES_DIR, "male_samuel_j.mp3"),
        'A confident announcer speaks proudly, "And now, the moment you have all been waiting for." '
        'He chuckles knowingly, "Heheh, trust me, this one is going to blow you away."',
    ),
    # ── Long-form examples (~30 s each) ───────────────────────────────────────
    # These pair a richer multi-beat scene with gen_duration = 30 s in the
    # Examples row below so the model is asked for a full half-minute clip.
    (
        "30s • Villain soliloquy",
        os.path.join(_VOICES_DIR, "male_harvey_keitel.mp3"),
        'A shadowy villain stands at the edge of his throne room, gazing into the dark. '
        'He speaks with slow, measured menace, "So, the little hero has come to finish me, has he?" '
        'He chuckles low and humourless, "Hehe, oh how delightfully predictable you mortals are." '
        'His voice hardens into ice, "I have lived ten thousand years. I have seen empires rise and fall like the tide." '
        'He scoffs, "And you think you, with your borrowed sword and your trembling hands, will be the one to end me?" '
        'A long pause. He whispers, almost tenderly, "I will give you a single chance to turn around and walk away." '
        'Then his voice rises with crushing finality, "Choose, child. The door behind you, or the grave at your feet."',
    ),
    (
        "30s • Late-night radio monologue",
        os.path.join(_VOICES_DIR, "male_old_movie.wav"),
        'A radio host clears his throat softly into the microphone in the late hours of the night. '
        'He settles into a warm, smoky tone, "Good evening, dear listeners, and welcome back to the After Hours Hour." '
        'He sighs contentedly, "Mmm, what a night it has been. The rain is tapping at my window like an old friend." '
        'He chuckles softly, "Heheh, you know the kind of friend, the one that always shows up unannounced." '
        'His voice drops, intimate, "I want you to lean back, wherever you are. Pour yourself something warm." '
        'He pauses, breath audible, "Tonight we are going to talk about love, and loss, and the songs that hold us together." '
        'A smile in his voice, "And I have got the perfect record cued up to start us off, so stay right where you are."',
    ),
    (
        "30s • Stand-up wheeze-laugh",
        os.path.join(_VOICES_DIR, "male_conan.mp3"),
        'A talk show host walks out and the crowd is already roaring. He gasps in mock outrage, "No! No no no!" '
        'He bursts into uncontrollable laughter, "Hahahaha, oh my god, oh my god, you cannot do that to me already!" '
        'He wheezes, gasping for air, "I have not even, hahaha, I have not even said hello yet!" '
        'He tries to compose himself, "Okay, okay, just give me a second here, give me a second." '
        'He clears his throat dramatically, "Ahem. Good evening, ladies and gentlemen." '
        'Then he loses it again, "Hahaha! No, sorry, sorry, I just remembered what happened in the green room." '
        'He pants, "Oh man, oh man, this is going to be one of those nights, I can already tell."',
    ),
    (
        "30s • Bedtime story",
        os.path.join(_VOICES_DIR, "female_shadowheart.wav"),
        'A mother sits at the edge of her child\'s bed in the dim glow of a single lamp. '
        'She speaks softly, "Once upon a time, in a kingdom by the sea, there lived a small dragon named Pip." '
        'She lowers her voice playfully, "Now Pip was not like the other dragons. Pip was afraid of fire." '
        'She smiles warmly, "Mmm, can you imagine? A dragon who was afraid of his own breath?" '
        'A gentle pause, "But Pip had something the other dragons did not have. Pip had courage in his heart." '
        'She hums softly, "Mmmmm. And one cold winter night, when the village below ran out of warmth..." '
        'Her voice drops to a whisper, "Pip closed his eyes, took a deep, deep breath, and remembered who he was."',
    ),
    (
        "30s • Sports commentary",
        os.path.join(_VOICES_DIR, "male_samuel_j.mp3"),
        'A sports commentator leans into the microphone with the crowd roaring around him. '
        'He shouts with rising energy, "Oh, this is it! This is the moment we have been waiting for all season!" '
        'He pants between phrases, "She has the ball at midfield, she is dribbling past one, past two!" '
        'A sudden gasp, "Oh my, what a move! Did you see that footwork, ladies and gentlemen?" '
        'His voice climbs, "She is twenty yards out, fifteen yards out, she winds back, and she SHOOTS!" '
        'A massive pause, then, "GOAAAAAAL! What a strike! What an absolute thunderbolt of a goal!" '
        'He laughs in disbelief, "Hahaha! Unbelievable! Forty thousand fans on their feet, and so am I!"',
    ),
]


app = Server()

# Serve static voice files and images
app.mount("/assets", StaticFiles(directory="assets"), name="assets")


@app.get("/", response_class=HTMLResponse)
async def homepage():
    html_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), "index.html")
    with open(html_path, "r", encoding="utf-8") as f:
        return f.read()


@app.api()
@spaces.GPU(duration=60)
def generate_audio(
    prompt: str,
    audio_ref: FileData | None,
    cfg: float,
    stg: float,
    dur_mult: float,
    gen_dur: float,
    ref_dur: float,
    seed: int
) -> FileData:
    if not prompt or not prompt.strip():
        raise gr.Error("Prompt is empty.")
    
    t0 = time.time()
    ref_path = None
    if audio_ref:
        if isinstance(audio_ref, dict):
            ref_path = audio_ref.get("path")
        elif hasattr(audio_ref, "path"):
            ref_path = audio_ref.path
    if ref_path and not os.path.exists(ref_path):
        ref_path = None
    
    output = tempfile.mktemp(suffix=".wav", prefix="dramabox_")
    tts.generate_to_file(
        prompt=prompt,
        output=output,
        voice_ref=ref_path,
        cfg_scale=cfg,
        stg_scale=stg,
        duration_multiplier=dur_mult,
        seed=int(seed),
        gen_duration=float(gen_dur),
        ref_duration=float(ref_dur),
    )
    elapsed = time.time() - t0
    logging.info(f"Generated in {elapsed:.2f}s -> {output}")
    return FileData(path=output)


if __name__ == "__main__":
    port = int(os.environ.get("GRADIO_SERVER_PORT", "7860"))
    app.launch(
        server_name="0.0.0.0",
        server_port=port,
        show_error=True
    )