app.py

import os
import gradio as gr
import soundfile as sf
import spaces
from huggingface_hub import snapshot_download

MODEL_REPO = "LyngualLabs/YorubaEnglish-CodeSwitching-TTS"
DEFAULT_REF_WAV = "reference.wav"
DEFAULT_REF_TEXT = "Ìròyìn ti sọ pé the government will ensure electricity tariff goes down ní January."

print("Downloading model...", flush=True)
model_dir = snapshot_download(MODEL_REPO)

from voxcpm import VoxCPM  # noqa: E402  (import after snapshot_download so the repo is cached)

print("Loading model...", flush=True)
model = VoxCPM.from_pretrained(model_dir, load_denoiser=False)
SR = model.tts_model.sample_rate
print("Model loaded.", flush=True)


@spaces.GPU(duration=120)
def synthesize(text, ref_audio, ref_text, cfg_value, inference_timesteps):
    if not text or not text.strip():
        raise gr.Error("Please enter some text to synthesize.")

    prompt_wav_path = ref_audio if ref_audio else DEFAULT_REF_WAV
    prompt_text = ref_text.strip() if (ref_audio and ref_text and ref_text.strip()) else (
        DEFAULT_REF_TEXT if not ref_audio else ""
    )
    if ref_audio and not prompt_text:
        raise gr.Error("Please provide the exact transcript of your uploaded reference audio.")

    wav = model.generate(
        text=text.strip(),
        prompt_wav_path=prompt_wav_path,
        prompt_text=prompt_text,
        cfg_value=cfg_value,
        inference_timesteps=int(inference_timesteps),
    )
    out_path = "output.wav"
    sf.write(out_path, wav, SR)
    return out_path


EXAMPLES = [
    ["Mo ní meeting pẹ̀lú marketing team ní aago mẹ́wàá lónìí, nítorí náà ẹ jẹ́ ká yára."],
    ["Ìròyìn ti sọ pé the government will ensure electricity tariff goes down ní January."],
    ["Wọ́n ní ìpàdé ní hotel tó wà ní Victoria Island, kì í ṣe ní office wa."],
]

CSS = """
@import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;600;700&display=swap');
:root, .dark { --tts-primary: #059669; --tts-primary-hover: #047857; --tts-accent: #b45309; }
.gradio-container { max-width: 960px !important; margin: auto !important;
                     font-family: 'Inter', system-ui, sans-serif !important; }
#header { text-align: center; padding: 0.5rem 0 0.25rem 0; }
#header h1 { font-size: 2.1rem; margin-bottom: 0.2rem; }
#subtitle { text-align: center; color: var(--body-text-color-subdued); font-size: 1.05rem;
            max-width: 640px; margin: 0 auto 0.75rem auto; }
#badges { display: flex; gap: 0.5rem; justify-content: center; flex-wrap: wrap; margin-bottom: 0.5rem; }
#generate-btn { font-size: 1.05rem !important; font-weight: 600 !important;
                background: var(--tts-primary) !important; border-color: var(--tts-primary) !important; }
#generate-btn:hover { background: var(--tts-primary-hover) !important; }
#footer { text-align: center; color: var(--body-text-color-subdued); font-size: 0.85rem;
          margin-top: 1rem; padding-top: 0.75rem; border-top: 1px solid var(--border-color-primary); }
"""

with gr.Blocks(title="Yoruba-English Code-Switching TTS", css=CSS) as demo:
    gr.Markdown("# 🎙️ Yoruba–English Code-Switching TTS", elem_id="header")
    gr.Markdown(
        "A voice-cloning text-to-speech model for Yoruba, including natural "
        "Yoruba–English code-switching — full fine-tuned from "
        "[VoxCPM2](https://huggingface.co/openbmb/VoxCPM2) on ~1,039 hours pooled from "
        "DSN African Voices, NaijaVoices, YECS, and WAXAL.",
        elem_id="subtitle",
    )
    gr.HTML(
        """
        <div id="badges">
            <a href="https://huggingface.co/LyngualLabs/YorubaEnglish-CodeSwitching-TTS" target="_blank">
                <img src="https://img.shields.io/badge/🤗%20Model-LyngualLabs-2e7d32?style=flat-square" />
            </a>
            <a href="https://huggingface.co/datasets/Professor/yoruba-speech-data" target="_blank">
                <img src="https://img.shields.io/badge/📊%20Dataset-1039h-b45309?style=flat-square" />
            </a>
        </div>
        """
    )
    gr.Markdown(
        "> ⚠️ **Experimental** — trained for only ~1 epoch. Quality varies by sentence; "
        "rarer words/proper nouns are more likely to be mispronounced."
    )

    with gr.Row(equal_height=False):
        with gr.Column(scale=1):
            text_in = gr.Textbox(
                label="Text to speak",
                placeholder="Type Yoruba, English, or mixed Yoruba-English text...",
                lines=4,
            )
            with gr.Accordion("🎭 Voice cloning — upload your own reference", open=False):
                ref_audio = gr.Audio(label="Reference audio (3-10s, clean)", type="filepath")
                ref_text = gr.Textbox(label="Exact transcript of the reference audio")
                gr.Markdown("_Leave blank to use the default built-in voice._")
            with gr.Accordion("⚙️ Advanced settings", open=False):
                cfg_value = gr.Slider(1.0, 4.0, value=2.0, step=0.1, label="CFG value (higher = sticks closer to reference)")
                inference_timesteps = gr.Slider(10, 35, value=22, step=1, label="Inference timesteps (higher = smoother, slower)")
            btn = gr.Button("✨ Generate Speech", variant="primary", elem_id="generate-btn", size="lg")
            gr.Examples(examples=EXAMPLES, inputs=[text_in], label="Try an example")
        with gr.Column(scale=1):
            audio_out = gr.Audio(label="Generated speech", type="filepath")

    gr.Markdown(
        "Part of the YECS-Downstream speech research effort by LyngualLabs · "
        "[Model card](https://huggingface.co/LyngualLabs/YorubaEnglish-CodeSwitching-TTS)",
        elem_id="footer",
    )

    btn.click(
        synthesize,
        inputs=[text_in, ref_audio, ref_text, cfg_value, inference_timesteps],
        outputs=[audio_out],
    )

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