mi@0: # Copyright 2014 Jason Heeris, jason.heeris@gmail.com
mi@0: # 
mi@0: # This file is part of the gammatone toolkit, and is licensed under the 3-clause
mi@0: # BSD license: https://github.com/detly/gammatone/blob/master/COPYING
mi@0: from __future__ import division
mi@0: import numpy as np
mi@0: 
mi@0: from .filters import make_erb_filters, centre_freqs, erb_filterbank
mi@0: 
mi@0: """
mi@0: This module contains functions for rendering "spectrograms" which use gammatone
mi@0: filterbanks instead of Fourier transforms.
mi@0: """
mi@0: '''
mi@0: Modified from the original toolbox. Jan 2015. --Mi
mi@0: 
mi@0: '''
mi@0: 
mi@0: def round_half_away_from_zero(num):
mi@0: 	""" Implement the round-half-away-from-zero rule, where fractional parts of
mi@0: 	0.5 result in rounding up to the nearest positive integer for positive
mi@0: 	numbers, and down to the nearest negative number for negative integers.
mi@0: 	"""
mi@0: 	return np.sign(num) * np.floor(np.abs(num) + 0.5)
mi@0: 
mi@0: 
mi@0: def gtgram_strides(fs, gammatoneLen, step_rate, filterbank_cols):
mi@0: 	"""
mi@0: 	Calculates the window size for a gammatonegram.
mi@0: 	
mi@0: 	@return a tuple of (window_size, hop_samples, output_columns)
mi@0: 	"""
mi@0: 	step_samples = int(gammatoneLen * step_rate)
mi@0: 	columns = (1 + int(np.floor((filterbank_cols - gammatoneLen)/ step_samples)))
mi@0: 		
mi@0: 	return (step_samples, columns)
mi@0: 
mi@0: 
mi@0: def gtgram_xe(wave, fs, f_max, channels, f_min):
mi@0: 	""" Calculate the intermediate ERB filterbank processed matrix """
mi@0: 	cfs = centre_freqs(f_max, channels, f_min)
mi@0: 	fcoefs = np.flipud(make_erb_filters(fs, cfs))
mi@0: 	xf = erb_filterbank(wave, fcoefs)
mi@0: 	return xf
mi@0: 	# xe = np.power(xf, 2)
mi@0: 	# return xe
mi@0: 
mi@0: 
mi@0: def gtgram(wave, fs, gammatoneLen, step_rate, channels, f_max, f_min):
mi@0: 	"""
mi@0: 	Calculate a spectrogram-like time frequency magnitude array based on
mi@0: 	gammatone subband filters. The waveform ``wave`` (at sample rate ``fs``) is
mi@0: 	passed through an multi-channel gammatone auditory model filterbank, with
mi@0: 	lowest frequency ``f_min`` and highest frequency ``f_max``. The outputs of
mi@0: 	each band then have their energy integrated over windows of ``window_time``
mi@0: 	seconds, advancing by ``hop_time`` secs for successive columns. These
mi@0: 	magnitudes are returned as a nonnegative real matrix with ``channels`` rows.
mi@0: 	
mi@0: 	| 2009-02-23 Dan Ellis dpwe@ee.columbia.edu
mi@0: 	|
mi@0: 	| (c) 2013 Jason Heeris (Python implementation)
mi@0: 	"""
mi@0: 	xe = gtgram_xe(wave, fs, f_max, channels, f_min)	 
mi@0: 	# print 'xe', xe.shape
mi@0: 	step_samples, ncols = gtgram_strides(fs, gammatoneLen, step_rate, xe.shape[1])
mi@0: 	# print gammatoneLen, step_samples, channels, ncols
mi@0: 	y = np.zeros((channels, ncols))
mi@0: 	
mi@0: 	for cnum in range(ncols):
mi@0: 		segment = xe[:, cnum * step_samples + np.arange(gammatoneLen)]
mi@0: 		segment = np.power(segment, 2)
mi@0: 		y[:, cnum] = np.sqrt(segment.mean(1))
mi@0: 	
mi@0: 	return y.T