app/services/audio_loader.py

import io
import numpy as np
import librosa

TARGET_SR = 22050


def load_audio_bytes(data: bytes) -> tuple[np.ndarray, int, float]:
    """Load raw audio bytes → (y, sr, duration_seconds). Resamples to 22050 Hz mono."""
    y, sr = librosa.load(io.BytesIO(data), sr=TARGET_SR, mono=True)
    return y, sr, float(len(y) / sr)


def chunk_audio(
    y: np.ndarray,
    sr: int,
    chunk_duration: float = 10.0,
    overlap: float = 0.5,
) -> list[dict]:
    """Split audio into overlapping chunks. Returns list of {y, start, end}."""
    chunk_samples = int(chunk_duration * sr)
    hop_samples   = int(chunk_samples * (1 - overlap))

    if len(y) <= chunk_samples:
        return [{"y": y, "start": 0.0, "end": float(len(y) / sr)}]

    chunks = []
    for s in range(0, len(y) - chunk_samples + 1, hop_samples):
        chunks.append({"y": y[s: s + chunk_samples], "start": s / sr, "end": (s + chunk_samples) / sr})
    return chunks