import numpy as np
import os
from scipy.io.wavfile import write as audio_write
### Generate data
data = np.random.uniform(size=(10000)) # single example
DATA = np.random.uniform(size=(10, 10000)) # multi example
wavfiles, numpyfiles = [], []
datafolder = 'data_intro/data'
os.makedirs(datafolder, exist_ok=True)
os.makedirs(datafolder + '_numpy', exist_ok=True)
for k, D in enumerate(DATA):
wavfiles.append(os.path.join(datafolder, str(k) + '.wav'))
numpyfiles.append(os.path.join(datafolder + '_numpy', str(k) + '.npy'))
np.save(numpyfiles[k], D)
audio_write(wavfiles[k], rate=1, data=D)
# -------------------------------------------------------------------------
### Create an STFT, get mean and std over time
from dabstract.dataprocessor import ProcessingChain
from dabstract.dataprocessor.processors import *
# create processing chain
dp = ProcessingChain()
dp.add(Framing(windowsize=10, stepsize=10, axis=0))
dp.add(FFT(axis=1))
dp.add(Aggregation(methods=['mean', 'std'], axis=0, combine='concatenate'))
dp.summary()
# apply processing chain to data
# make sure to provide sampling frequency to dp. Kwargs are always accessible for
# all processing layer. Therefore, you should make sure naming DOES NOT overlap
import sounddevice as sound_device
from scipy.io.wavfile import write as audio_write
fs=44100
seconds=15
recording=sound_device.rec(int(seconds*fs),samplerate=fs,channels=2)
sound_device.wait()
audio_write("recording.wav",fs,recording)