audio_utils.py

import librosa
import numpy as np
import torch
from config import CONFIG

def preprocess_audio(path, device):
    y, _ = librosa.load(path, sr=CONFIG["sample_rate"])

    max_len = int(CONFIG["sample_rate"] * CONFIG["duration"])
    y = y[:max_len] if len(y) > max_len else np.pad(y, (0, max_len - len(y)))

    mel = librosa.feature.melspectrogram(
        y=y,
        sr=CONFIG["sample_rate"],
        n_fft=CONFIG["n_fft"],
        hop_length=CONFIG["hop_length"],
        n_mels=CONFIG["n_mels"]
    )

    mel_db = librosa.power_to_db(mel, ref=np.max)
    mel_db = (mel_db - mel_db.mean()) / (mel_db.std() + 1e-9)

    tensor = torch.from_numpy(mel_db).unsqueeze(0).unsqueeze(0)
    return tensor.to(device)