utils.py

import os, requests
import matplotlib.pyplot as plt
import numpy as np
import librosa
import noisereduce as nr
from scipy.signal import butter, lfilter


def bandpass_filter(data, lowcut, highcut, sr, order=5):
    nyquist = 0.5 * sr
    low = lowcut / nyquist
    high = highcut / nyquist
    b, a = butter(order, [low, high], btype='band')
    y = lfilter(b, a, data)
    return y


def plot_mel(sr, x):
    mel_spec = librosa.feature.melspectrogram(y=x, sr=sr, n_mels=128, fmax=10000)
    mel_spec_db = librosa.power_to_db(mel_spec, ref=np.max)
    mel_spec_db = (mel_spec_db - mel_spec_db.min()) / (mel_spec_db.max() - mel_spec_db.min()) # normalize spectrogram to [0,1]
    # mel_spec_db = np.stack([mel_spec_db, mel_spec_db, mel_spec_db], axis=-1)  # Convert to 3-channel
    fig, ax = plt.subplots(nrows=1, ncols=1, sharex=True)
    librosa.display.specshow(mel_spec_db, sr=sr, x_axis='time', y_axis='mel', fmin = 0, fmax=10000, ax = ax)
    return fig


def plot_wave(sr, x):
    ry = nr.reduce_noise(y=x, sr=sr)
    fig, ax = plt.subplots(2, 1, figsize=(12, 8))

    # Plot the original waveform
    librosa.display.waveshow(x, sr=sr, ax=ax[0])
    ax[0].set(title='Original Waveform')
    ax[0].set_xlabel('Time (s)')
    ax[0].set_ylabel('Amplitude')

    # Plot the noise-reduced waveform
    librosa.display.waveshow(ry, sr=sr, ax=ax[1])
    ax[1].set(title='Noise Reduced Waveform')
    ax[1].set_xlabel('Time (s)')
    ax[1].set_ylabel('Amplitude')

    plt.tight_layout()
    return fig


def download_model(url, model_path):
    response = requests.get(url)
    response.raise_for_status()  # Ensure the request was successful
    with open(model_path, 'wb') as f:
        f.write(response.content)