pyACA.ToolMfccFb Namespace Reference

Functions
	ToolMfccFb (iFftLength, f_s)
	helper function: computes transfer functions of MFCC filter bands see function from Slaneys Auditory Toolbox (Matlab)

Function Documentation

ToolMfccFb()

ToolMfccFb	(		iFftLength,
			f_s )

helper function: computes transfer functions of MFCC filter bands see function from Slaneys Auditory Toolbox (Matlab)

Parameters

iFftLength	length of FFT
f_s	sample rate of audio data

Returns

H: matrix with transfer functions

Definition at line 13 of file ToolMfccFb.py.

def ToolMfccFb(iFftLength, f_s):
 
    # initialization
    f_start = 133.3333
 
    iNumLinFilters = 13
    iNumLogFilters = 27
    iNumFilters = iNumLinFilters + iNumLogFilters
 
    linearSpacing = 66.66666666
    logSpacing = 1.0711703
 
    # compute band frequencies
    f = np.zeros(iNumFilters + 2)
    f[np.arange(0, iNumLinFilters)] = f_start + np.arange(0, iNumLinFilters) * linearSpacing
    f[np.arange(iNumLinFilters, iNumFilters + 2)] = f[iNumLinFilters - 1] * logSpacing**np.arange(1, iNumLogFilters + 3)
 
    # sanity check
    if f[iNumFilters - 1] >= f_s / 2:
        f = f[f < f_s / 2]
        iNumFilters = f.shape[0] - 2
 
    f_l = f[np.arange(0, iNumFilters)]
    f_c = f[np.arange(1, iNumFilters + 1)]
    f_u = f[np.arange(2, iNumFilters + 2)]
 
    # allocate memory for filters and set max amplitude
    H = np.zeros([iNumFilters, iFftLength])
    afFilterMax = 2 / (f_u - f_l)
    f_k = np.arange(0, iFftLength) / (iFftLength - 1) * f_s / 2
 
    # compute the transfer functions
    for c in range(0, iNumFilters):
        # lower filter slope
        i_l = np.argmax(f_k > f_l[c])
        i_u = np.max([0, np.argmin(f_k <= f_c[c]) - 1])
        H[c, np.arange(i_l, i_u + 1)] = afFilterMax[c] * (f_k[np.arange(i_l, i_u + 1)] - f_l[c]) / (f_c[c] - f_l[c])
        # upper filter slope
        i_l = i_u + 1
        i_u = np.max([0, np.argmin(f_k < f_u[c]) - 1])
        H[c, np.arange(i_l, i_u + 1)] = afFilterMax[c] * (f_u[c] - f_k[np.arange(i_l, i_u + 1)]) / (f_u[c] - f_c[c])
 
    return H