preprocessing.py

import librosa
import numpy as np

TARGET_SR = 16000

def load_audio(file_path):

    audio, sr = librosa.load(file_path, sr=TARGET_SR)

    return audio


def normalize_audio(audio):

    audio = audio / np.max(np.abs(audio))

    return audio


def remove_silence(audio):

    intervals = librosa.effects.split(audio, top_db=30)

    segments = []

    for start, end in intervals:
        segments.append(audio[start:end])

    if len(segments) == 0:
        return audio

    return np.concatenate(segments)