kokoro_tts.py

from kokoro import KPipeline
import soundfile as sf
import numpy as np
import logging


def generate_audio(
    text,
    lang_code="a",
    voice="af_heart",
    speed=1,
    save_segments=False,
    progress=None,
):
    """
    Generate audio from text using Kokoro TTS pipeline

    Args:
        text (str): Text to convert to speech
        lang_code (str): Language code for the TTS model
        voice (str): Voice ID to use
        speed (float): Speech speed multiplier
        save_segments (bool): Whether to save individual audio segments

    Returns:
        numpy.ndarray: Combined audio data at 24kHz sample rate
    """
    pipeline = KPipeline(lang_code=lang_code)

    generator = pipeline(text, voice=voice, speed=speed, split_pattern=r"\.")

    all_audio = []
    segments = list(generator)  # Get total number of segments

    for i, (gs, ps, audio) in enumerate(
        progress.tqdm(segments, desc="Generating audio")
    ):
        logging.info("Processing segment")
        logging.info(f"Graphemes: {gs}")
        logging.info(f"Phonemes: {ps}")
        all_audio.append(audio)

        if save_segments:
            sf.write(f"segment_{i}.wav", audio, 24000)

    # Concatenate all audio segments
    combined_audio = np.concatenate(all_audio)
    return combined_audio


if __name__ == "__main__":
    # Example usage
    sample_text = "Hello world"
    audio_data = generate_audio(sample_text)
    sf.write("out.wav", audio_data, 24000)