import random
import subprocess
import numpy as np
from scipy.io.wavfile import read


def weights_init(m):
    classname = m.__class__.__name__
    if classname.find("Conv") != -1:
        m.weight.data.normal_(0.0, 0.02)
    elif classname.find("BatchNorm2d") != -1:
        m.weight.data.normal_(1.0, 0.02)
        m.bias.data.fill_(0)

def get_commit_hash():
    message = subprocess.check_output(["git", "rev-parse", "--short", "HEAD"])
    return message.strip().decode('utf-8')

def read_wav_np(path):
    sr, wav = read(path)

    if len(wav.shape) == 2:
        wav = wav[:, 0]

    if wav.dtype == np.int16:
        wav = wav / 32768.0
    elif wav.dtype == np.int32:
        wav = wav / 2147483648.0
    elif wav.dtype == np.uint8:
        wav = (wav - 128) / 128.0

    wav = wav.astype(np.float32)

    return sr, wav