igammatone

class diffsptk.GammatoneFilterBankSynthesis(sample_rate, *, desired_delay=4, f_min=70, f_base=1000, f_max=6700, filter_order=4, bandwidth_factor=1.0, density=1.0, exact=False, n_iter=100, eps=1e-08)

Gammatone filter bank analysis.

Parameters:

sample_rateint >= 1

Sample rate in Hz.

desired_delayfloat > 0

Desired delay in milliseconds.

f_minfloat >= 0

Minimum frequency in Hz.

f_basefloat >= 0

Base frequency in Hz.

f_maxfloat <= sample_rate // 2

Maximum frequency in Hz.

filter_orderint >= 1

Order of Gammatone filter.

bandwidth_factorfloat > 0

Bandwidth of Gammatone filter.

densityfloat > 0

Density of frequencies on the ERB scale.

exactbool

If False, use all-pole approximation.

n_iterint >= 1

Number of iterations for gain computation.

epsfloat >= 0

Tolerance for gain computation.

References

[1]

V. Hohmann, “Frequency analysis and synthesis using a Gammatone filterbank,” Acta Acustica united with Acustica, vol. 88, no. 3, pp. 433-442, 2002.

[2]

T. Herzke, “Improved numerical methods for Gammatone filterbank analysis and synthesis,” Acta Acustica united with Acustica, vol. 93, no. 3, pp. 498-500, 2007.

forward(y, keepdim=True, compensate_delay=True)

Reconstruct waveform from filter bank signals.

Parameters:

yTensor [shape=(B, K, T) or (K, T)]

Filtered signals.

keepdimbool

If True, the output shape is (B, 1, T) instead (B, T).

compensate_delaybool

If True, compensate the delay.

Returns:

outTensor [shape=(B, 1, T) or (B, T)]

Reconstructed waveform.

Examples

>>> x = diffsptk.impulse(15999)
>>> x[:5]
tensor([1., 0., 0., 0., 0.])
>>> f = diffsptk.GammatoneFilterBankAnalysis(16000)
>>> g = diffsptk.GammatoneFilterBankSynthesis(16000)
>>> y = g(f(x)).squeeze()
>>> y[:5]
tensor([ 0.8349,  0.0682, -0.1085,  0.0559, -0.0947])

See also

gammatone