visualization.py

from __future__ import print_function
from __future__ import division
import time
import numpy as np
import state
import dsp

# Handler for audio samples. Will be updated in 'on_state_change'
process_sample = lambda samples, logger: None

def on_state_change (config, visualization):
    print('on_state_change')
    global process_sample
    # FFT window shape
    fft_window = np.hanning(config['fft_samples_per_window'])

    # FFT binning (mel-bank-transformation)
    mel_trafo, (mel_x, fft_x) = dsp.compute_melmat(
        num_mel_bands=config['FFT_N_BINS'],
        freq_min=config['MIN_FREQUENCY'],
        freq_max=config['MAX_FREQUENCY'],
        num_fft_bands=int(config['fft_samples_per_window'] / 2),
        sample_rate=config['SAMPLE_RATE']
    )

    #Smoothing on spectrum output
    mel_smoothing = dsp.ExpFilter(np.tile(1e-1, config['FFT_N_BINS']), alpha_decay=0.8, alpha_rise=0.99)

    # Audio sample rolling window
    y_roll = np.random.rand(config['fft_samples_per_window']) / 1e16

    def update(audio_samples, logger):
        nonlocal y_roll
        logger('idle')

        # Normalize samples between 0 and 1
        y_update = audio_samples / 2.0**15 + 0.5
        # Construct a rolling window of audio samples
        l = len(y_update)
        y_roll[:-l] = y_roll[l:]
        y_roll[-l:] = y_update
        y_data = y_roll.astype(np.float32)

        logger('roll')

        # Fourier transform and mel transformation
        N = len(y_data)
        fft = np.abs(np.fft.rfft(y_data * fft_window)[:N // 2]) #np.zeros(N//2)#
        mel = mel_trafo(fft)
        # mel = mel**2

        logger('fft')
        
        # Visualize
        # mel = mel_smoothing.update(mel)
        led_output = visualization(mel, y_data, (fft, fft_x))

        logger('vis')

        return (y_data, mel, led_output)

    # Update sample handler
    process_sample = update

state.register_on_state_change_hander(on_state_change)