audio_utils.py

import librosa
import numpy as np
import cv2

SR = 16000
DURATION = 4.0
N_MELS = 192
N_FFT = 2048
HOP_LENGTH = 160
IMG_SIZE = 224


def audio_to_spectrogram(wav_path):

    y, _ = librosa.load(wav_path, sr=SR)
    y, _ = librosa.effects.trim(y, top_db=30)

    target = int(SR * DURATION)

    if len(y) < target:
        pad = target - len(y)
        y = np.pad(y, (pad // 2, pad - pad // 2))
    else:
        y = y[:target]

    mel = librosa.feature.melspectrogram(
        y=y,
        sr=SR,
        n_fft=N_FFT,
        hop_length=HOP_LENGTH,
        n_mels=N_MELS
    )

    logmel = librosa.power_to_db(mel, ref=np.max)

    logmel = (logmel - logmel.min()) / (logmel.max() - logmel.min())

    img = (logmel * 255).astype(np.uint8)
    img = cv2.resize(img, (IMG_SIZE, IMG_SIZE))
    img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)

    return img