app.py

#!/usr/bin/env python3
"""
Nigerian TTS API - YarnGPT-based TTS for Nigerian languages.
Runs on HuggingFace Spaces with T4 GPU.
"""

import io
import logging
import time
import tempfile
import torch
import torchaudio
import gradio as gr

logging.basicConfig(level=logging.INFO)
log = logging.getLogger(__name__)

# Check GPU
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
log.info(f"Device: {DEVICE}")
if DEVICE == "cuda":
    log.info(f"GPU: {torch.cuda.get_device_name(0)}")


def get_system_info():
    """Return system information."""
    info = f"Device: {DEVICE}\n"
    info += f"PyTorch: {torch.__version__}\n"
    if DEVICE == "cuda":
        info += f"GPU: {torch.cuda.get_device_name(0)}\n"
        mem = torch.cuda.get_device_properties(0).total_memory / 1e9
        info += f"Memory: {mem:.1f} GB"
    return info


def tts_synthesize(text: str, speaker: str, language: str):
    """TTS synthesis using YarnGPT."""
    if not text.strip():
        return None, "Empty text"

    log.info(f"TTS: text='{text[:50]}...', speaker={speaker}, lang={language}")
    start = time.time()

    try:
        from yarngpt import generate_speech

        # Generate speech
        audio_tensor = generate_speech(
            text=text,
            speaker=speaker,
            language=language,
            temperature=0.1,
            repetition_penalty=1.1,
            max_length=4000,
        )

        elapsed = time.time() - start
        log.info(f"TTS done in {elapsed:.1f}s")

        # Save to temp file
        with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as f:
            torchaudio.save(f.name, audio_tensor, sample_rate=24000)
            return f.name, None

    except Exception as e:
        log.error(f"TTS error: {e}")
        import traceback
        traceback.print_exc()
        return None, str(e)


# Available speakers by language
SPEAKERS = {
    "english": ["idera", "chinenye", "jude", "emma", "umar", "joke", "zainab", "osagie", "remi", "tayo"],
    "yoruba": ["abayomi", "aisha", "folake"],
    "igbo": ["chioma", "obinna", "adanna"],
    "hausa": ["amina", "fatima", "ibrahim", "yusuf"],
}

ALL_SPEAKERS = []
for speakers in SPEAKERS.values():
    ALL_SPEAKERS.extend(speakers)


def update_speakers(language):
    """Update speaker dropdown based on language."""
    speakers = SPEAKERS.get(language, SPEAKERS["english"])
    return gr.Dropdown(choices=speakers, value=speakers[0])


# Gradio UI
with gr.Blocks(title="Nigerian TTS API") as demo:
    gr.Markdown("# 🎙️ Nigerian TTS API")
    gr.Markdown("YarnGPT-based Text-to-Speech for Nigerian languages")

    with gr.Row():
        with gr.Column():
            text_input = gr.Textbox(
                label="Text",
                placeholder="Enter text to synthesize...",
                lines=3,
            )
            language = gr.Dropdown(
                label="Language",
                choices=["english", "yoruba", "igbo", "hausa"],
                value="english",
            )
            speaker = gr.Dropdown(
                label="Speaker",
                choices=SPEAKERS["english"],
                value="idera",
            )
            submit_btn = gr.Button("🔊 Synthesize", variant="primary")

        with gr.Column():
            audio_output = gr.Audio(label="Output Audio", type="filepath")
            error_output = gr.Textbox(label="Status", visible=True)
            sys_info = gr.Textbox(label="System Info", value=get_system_info(), lines=4)

    # Update speakers when language changes
    language.change(fn=update_speakers, inputs=[language], outputs=[speaker])

    # Generate speech on button click
    submit_btn.click(
        fn=tts_synthesize,
        inputs=[text_input, speaker, language],
        outputs=[audio_output, error_output],
    )

    # Example inputs
    gr.Examples(
        examples=[
            ["Hello, how are you today?", "idera", "english"],
            ["The weather in Lagos is beautiful.", "emma", "english"],
            ["Bawo ni, e ku ojo.", "abayomi", "yoruba"],
            ["How you dey, my brother?", "jude", "english"],
        ],
        inputs=[text_input, speaker, language],
    )

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