# To install the library for the recognition and organization of speech and audio (librosa):
# pip install librosa
# https://github.com/librosa/librosa/blob/master/examples/LibROSA%20demo.ipynb

import librosa
import numpy as np
from io import TextIOWrapper
from zipfile import ZipFile

def makeTensors(partition, n):
    X = np.zeros((n, 32, 20)).astype("float32")
    with ZipFile("ml530-2022-fall-speech.zip", "r") as archive:
        for index in range(n):
            with archive.open(partition + "/" + partition + "/" + str(index).zfill(5) + ".wav") as file:
                timeSeries, samplingRate = librosa.load(file, sr = None)
                if (timeSeries.shape[0] < samplingRate):
                    timeSeries = librosa.effects.time_stretch(timeSeries, rate = 1.0 * timeSeries.shape[0] / samplingRate)
                length = timeSeries.shape[0]
                if (length > samplingRate):
                    length = samplingRate
                mfcc = librosa.feature.mfcc(y = timeSeries[:length], sr = samplingRate)
                X[index,:,:] = np.transpose(mfcc).astype("float32")
    np.save(partition + "X.npy", X)
    if (partition != "tst"):
        Y = np.zeros(n).astype("int32")
        with ZipFile("ml530-2022-fall-speech.zip", "r") as archive:
            with TextIOWrapper(archive.open(partition + ".csv", "r")) as file:
                header = file.readline()
                for index in range(n):
                    Y[index] = np.int32(file.readline().strip("\r\n").split(",")[1])
        np.save(partition + "Y.npy", Y)

makeTensors("trn", 51088)
makeTensors("val", 6798)
makeTensors("tst", 6835)