Source code for cnmodel.util.sound

"""
Tools for generating auditory stimuli. 
"""
from __future__ import division
import numpy as np
import scipy
import scipy.io.wavfile
import resampy

[docs]def create(type, **kwds): """ Create a Sound instance using a key returned by Sound.key(). """ cls = globals()[type] return cls(**kwds)
[docs]class Sound(object): """ Base class for all sound stimulus generators. """ def __init__(self, duration, rate=100e3, **kwds): """ Parameters ---------- duration: float (no default): duration of the stimulus, in seconds rate : float (default: 100000.) sample rate for sound generation """ self.opts = {'rate': rate, 'duration': duration} self.opts.update(kwds) self._time = None self._sound = None @property def sound(self): """ :obj:`array`: The generated sound array, expressed in Pascals. """ if self._sound is None: self._sound = self.generate() return self._sound @property def time(self): """ :obj:`array`: The array of time values, expressed in seconds. """ if self._time is None: self._time = np.linspace(0, self.opts['duration'], self.num_samples) return self._time @property def num_samples(self): """ int: The number of samples in the sound array. """ return 1 + int(self.opts['duration'] * self.opts['rate']) @property def dt(self): """ float: the sample period (time step between samples). """ return 1.0 / self.opts['rate'] @property def duration(self): """ float: The duration of the sound """ return self.opts['duration']
[docs] def key(self): """ The sound can be recreated using ``create(**key)``. :obj:`dict`: Return dict of parameters needed to completely describe this sound. """ k = self.opts.copy() k['type'] = self.__class__.__name__ return k
[docs] def measure_dbspl(self, tstart, tend): """ Measure the sound pressure for the waveform in a window of time Parameters ---------- tstart : time to start spl measurement (seconds). tend : ending time for spl measurement (seconds). Returns ------- float : The measured amplitude (dBSPL) of the sound from tstart to tend """ istart = int(tstart * self.opts['rate']) iend = int(tend * self.opts['rate']) return pa_to_dbspl(self.sound[istart:iend].std())
[docs] def generate(self): """ Generate and return the sound output. This method is defined by subclasses. """ raise NotImplementedError()
def __getattr__(self, name): if 'opts' not in self.__dict__: raise AttributeError(name) if name in self.opts: return self.opts[name] else: return object.__getattr__(self, name)
[docs]class TonePip(Sound): """ Create one or more tone pips with cosine-ramped edges. Parameters ---------- rate : float Sample rate in Hz duration : float Total duration of the sound f0 : float or array-like Tone frequency in Hz. Must be less than half of the sample rate. dbspl : float Maximum amplitude of tone in dB SPL. pip_duration : float Duration of each pip including ramp time. Must be at least 2 * ramp_duration. pip_start : array-like Start times of each pip ramp_duration : float Duration of a single ramp period (from minimum to maximum). This may not be more than half of pip_duration. """ def __init__(self, **kwds): reqdWords = ['rate', 'duration', 'f0', 'dbspl', 'pip_duration', 'pip_start', 'ramp_duration'] for k in reqdWords: if k not in kwds.keys(): raise TypeError("Missing required argument '%s'" % k) if kwds['pip_duration'] < kwds['ramp_duration'] * 2: raise ValueError("pip_duration must be greater than (2 * ramp_duration).") if kwds['f0'] > kwds['rate'] * 0.5: raise ValueError("f0 must be less than (0.5 * rate).") Sound.__init__(self, **kwds)
[docs] def generate(self): """ Call to compute the tone pips Returns ------- array : generated waveform """ o = self.opts return piptone(self.time, o['ramp_duration'], o['rate'], o['f0'], o['dbspl'], o['pip_duration'], o['pip_start'])
[docs]class FMSweep(Sound): """ Create an FM sweep with either linear or logarithmic rates, of a specified duration between two frequencies. Parameters ---------- rate : float Sample rate in Hz duration : float Total duration of the sweep start : float t times of each pip freqs : list [f0, f1]: the start and stop times for the sweep ramp : string valid input for type of sweep (linear, logarithmic, etc) dbspl : float Maximum amplitude of pip in dB SPL. """ def __init__(self, **kwds): for k in ['rate', 'duration', 'start', 'freqs', 'ramp', 'dbspl']: if k not in kwds: raise TypeError("Missing required argument '%s'" % k) Sound.__init__(self, **kwds)
[docs] def generate(self): """ Call to actually compute the the FM sweep Returns ------- array : generated waveform """ o = self.opts return fmsweep(self.time, o['start'], o['duration'], o['freqs'], o['ramp'], o['dbspl'])
[docs]class NoisePip(Sound): """ One or more noise pips with cosine-ramped edges. Parameters ---------- rate : float Sample rate in Hz duration : float Total duration of the sound seed : int >= 0 Random seed dbspl : float Maximum amplitude of tone in dB SPL. pip_duration : float Duration of each pip including ramp time. Must be at least 2 * ramp_duration. pip_start : array-like Start times of each pip ramp_duration : float Duration of a single ramp period (from minimum to maximum). This may not be more than half of pip_duration. """ def __init__(self, **kwds): for k in ['rate', 'duration', 'dbspl', 'pip_duration', 'pip_start', 'ramp_duration', 'seed']: if k not in kwds: raise TypeError("Missing required argument '%s'" % k) if kwds['pip_duration'] < kwds['ramp_duration'] * 2: raise ValueError("pip_duration must be greater than (2 * ramp_duration).") if kwds['seed'] < 0: raise ValueError("Random seed must be integer > 0") Sound.__init__(self, **kwds)
[docs] def generate(self): """ Call to compute the noise pips Returns ------- array : generated waveform """ o = self.opts return pipnoise(self.time, o['ramp_duration'], o['rate'], o['dbspl'], o['pip_duration'], o['pip_start'], o['seed'])
[docs]class ClickTrain(Sound): """ One or more clicks (rectangular pulses). Parameters ---------- rate : float Sample rate in Hz dbspl : float Maximum amplitude of click in dB SPL. click_duration : float Duration of each click including ramp time. Must be at least 1/rate. click_starts : array-like Start times of each click """ def __init__(self, **kwds): for k in ['rate', 'duration', 'dbspl', 'click_duration', 'click_starts']: if k not in kwds: raise TypeError("Missing required argument '%s'" % k) if kwds['click_duration'] < 1./kwds['rate']: raise ValueError("click_duration must be greater than sample rate.") Sound.__init__(self, **kwds)
[docs] def generate(self): o = self.opts return clicks(self.time, o['rate'], o['dbspl'], o['click_duration'], o['click_starts'])
class SAMNoise(Sound): """ One or more gaussian noise pips with cosine-ramped edges. Parameters ---------- rate : float Sample rate in Hz duration : float Total duration of the sound seed : int >= 0 Random seed dbspl : float Maximum amplitude of pip in dB SPL. pip_duration : float Duration of each pip including ramp time. Must be at least 2 * ramp_duration. pip_start : array-like Start times of each pip ramp_duration : float Duration of a single ramp period (from minimum to maximum). This may not be more than half of pip_duration. fmod : float SAM modulation frequency dmod : float Modulation depth """ def __init__(self, **kwds): parms = ['rate', 'duration', 'seed', 'pip_duration', 'pip_start', 'ramp_duration', 'fmod', 'dmod', 'seed'] for k in parms: if k not in kwds: raise TypeError("Missing required argument '%s'" % k) if kwds['pip_duration'] < kwds['ramp_duration'] * 2: raise ValueError("pip_duration must be greater than (2 * ramp_duration).") if kwds['seed'] < 0: raise ValueError("Random seed must be integer > 0") Sound.__init__(self, **kwds) def generate(self): """ Call to compute the SAM noise Returns ------- array : generated waveform """ o = self.opts o['phaseshift'] = 0. return modnoise(self.time, o['ramp_duration'], o['rate'], o['f0'], o['pip_duration'], o['pip_start'], o['dbspl'], o['fmod'], o['dmod'], 0., o['seed'])
[docs]class SAMTone(Sound): """ SAM tones with cosine-ramped edges. Parameters ---------- rate : float Sample rate in Hz duration : float Total duration of the sound f0 : float or array-like Tone frequency in Hz. Must be less than half of the sample rate. dbspl : float Maximum amplitude of tone in dB SPL. pip_duration : float Duration of each pip including ramp time. Must be at least 2 * ramp_duration. pip_start : array-like Start times of each pip ramp_duration : float Duration of a single ramp period (from minimum to maximum). This may not be more than half of pip_duration. fmod : float SAM modulation frequency, Hz dmod : float Modulation depth, % """ def __init__(self, **kwds): for k in ['rate', 'duration', 'f0', 'dbspl', 'pip_duration', 'pip_start', 'ramp_duration', 'fmod', 'dmod']: if k not in kwds: raise TypeError("Missing required argument '%s'" % k) if kwds['pip_duration'] < kwds['ramp_duration'] * 2: raise ValueError("pip_duration must be greater than (2 * ramp_duration).") if kwds['f0'] > kwds['rate'] * 0.5: raise ValueError("f0 must be less than (0.5 * rate).") Sound.__init__(self, **kwds)
[docs] def generate(self): """ Call to compute a SAM tone Returns ------- array : generated waveform """ o = self.opts basetone = piptone(self.time, o['ramp_duration'], o['rate'], o['f0'], o['dbspl'], o['pip_duration'], o['pip_start']) return sinusoidal_modulation(self.time, basetone, o['pip_start'], o['fmod'], o['dmod'], 0.)
[docs]def pa_to_dbspl(pa, ref=20e-6): """ Convert Pascals (rms) to dBSPL. By default, the reference pressure is 20 uPa. """ return 20 * np.log10(pa / ref)
[docs]def dbspl_to_pa(dbspl, ref=20e-6): """ Convert dBSPL to Pascals (rms). By default, the reference pressure is 20 uPa. """ return ref * 10**(dbspl / 20.0)
[docs]class SAMNoise(Sound): """ One or more gaussian noise pips with cosine-ramped edges, sinusoidally modulated. Parameters ---------- rate : float Sample rate in Hz duration : float Total duration of the sound seed : int >= 0 Random seed dbspl : float Maximum amplitude of pip in dB SPL. pip_duration : float Duration of each pip including ramp time. Must be at least 2 * ramp_duration. pip_start : array-like Start times of each pip ramp_duration : float Duration of a single ramp period (from minimum to maximum). This may not be more than half of pip_duration. fmod : float SAM modulation frequency dmod : float Modulation depth Returns ------- array : waveform """ def __init__(self, **kwds): for k in ['rate', 'duration', 'seed', 'pip_duration', 'pip_start', 'ramp_duration', 'fmod', 'dmod']: if k not in kwds: raise TypeError("Missing required argument '%s'" % k) if kwds['pip_duration'] < kwds['ramp_duration'] * 2: raise ValueError("pip_duration must be greater than (2 * ramp_duration).") if kwds['seed'] < 0: raise ValueError("Random seed must be integer > 0") Sound.__init__(self, **kwds)
[docs] def generate(self): o = self.opts basenoise = pipnoise(self.time, o['ramp_duration'], o['rate'], o['dbspl'], o['pip_duration'], o['pip_start'], o['seed']) return sinusoidal_modulation(self.time, basenoise, o['pip_start'], o['fmod'], o['dmod'], 0.)
[docs]class ReadWavefile(Sound): """ Read a .wav file from disk, possibly converting the sample rate and the scale for use in driving the auditory nerve fiber model. Parameters ---------- wavefile : str name of the .wav file to read rate : float Sample rate in Hz (waveform will be resampled to this rate) channel: int (default: 0) If wavefile has 2 channels, select 0 or 1 for the channel to read dbspl : float or None If specified, the wave file is scaled such that its overall dBSPL (measured from RMS of the entire waveform) is equal to this value. Either ``dbspl`` or ``scale`` must be specified. scale : float or None If specified, the wave data is multiplied by this value to yield values in dBSPL. Either ``dbspl`` or ``scale`` must be specified. delay: float (default: 0.) Silent delay time to start sound, in s. Allows anmodel and cells to run to steady-state maxdur : float or None (default: None) If specified, maxdur defines the total duration of generated waveform to return (in seconds). If None, the generated waveform duration will be the sum of any delay value and the duration of the waveform from the wavefile. Returns ------- array : waveform """ def __init__(self, wavefile, rate, channel=0, dbspl=None, scale=None, delay=0., maxdur=None): if dbspl is not None and scale is not None: raise ValueError('Only one of "dbspl" or "scale" can be set') duration = 0. # forced because of the way num_samples has to be calculated first if delay < 0.: raise ValueError('ReadWavefile: delay must be > 0., got: %f' % delay) if maxdur is not None and maxdur < 0: raise ValueError('ReadWavefile: maxdur must be None or > 0., got: %f' % maxdur) Sound.__init__(self, duration, rate, wavefile=wavefile, channel=channel, dbspl=dbspl, scale=scale, maxdur=maxdur, delay=delay)
[docs] def generate(self): """ Read the wave file from disk, clip duration, resample if necessary, and scale Returns ------- array : generated waveform """ [fs_wav, stimulus] = scipy.io.wavfile.read(self.opts['wavefile']) # raw is a numpy array of integer, representing the samples if len(stimulus.shape) > 1 and stimulus.shape[1] > 0: stimulus = stimulus[:,self.opts['channel']] # just use selected channel fs_wav = float(fs_wav) maxdur = self.opts['maxdur'] delay = self.opts['delay'] delay_array = np.zeros(int(delay*fs_wav)) # build delay array (may have 0 length) if maxdur is None: maxdur = delay + len(stimulus)/fs_wav # true total length maxpts = int(maxdur * fs_wav) stimulus = np.hstack((delay_array, stimulus))[:maxpts] if self.opts['rate'] != fs_wav: stimulus = resampy.resample(stimulus, fs_wav, self.opts['rate']) self.opts['duration'] = (stimulus.shape[0]-1)/self.opts['rate'] # compute the duration, match for linspace calculation used in time. self._time = None self.time # requesting time should cause recalulation of the time if self.opts['dbspl'] is not None: rms = np.sqrt(np.mean(stimulus**2.0)) # find rms of the waveform stimulus = dbspl_to_pa(self.opts['dbspl'] ) * stimulus / rms # scale into Pascals if self.opts['scale'] is not None: stimulus = stimulus * self.opts['scale'] return stimulus
[docs]def sinusoidal_modulation(t, basestim, tstart, fmod, dmod, phaseshift): """ Impose a sinusoidal amplitude-modulation on the input waveform. For dmod=100%, the envelope max is 2, the min is 0; for dmod = 0, the max and min are 1 maintains equal energy for all modulation depths. Equation from Rhode and Greenberg, J. Neurophys, 1994 (adding missing parenthesis) and Sayles et al. J. Physiol. 2013 The envelope can be phase shifted (useful for co-deviant stimuli). Parameters ---------- t : array array of waveform time values (seconds) basestim : array array of waveform values that will be subject to sinulsoidal envelope modulation tstart : float time at which the base sound starts (modulation starts then, with 0 phase crossing) (seconds) fmod : float modulation frequency (Hz) dmod : float modulation depth (percent) phaseshift : float modulation phase shift (starting phase, radians) """ env = (1.0 + (dmod/100.0) * np.sin((2.0*np.pi*fmod*(t-tstart)) + phaseshift - np.pi/2)) # envelope... return basestim*env
[docs]def modnoise(t, rt, Fs, F0, dur, start, dBSPL, FMod, DMod, phaseshift, seed): """ Generate an amplitude-modulated noise with linear ramps. Parameters ---------- t : array array of waveform time values rt : float ramp duration Fs : float sample rate F0 : float tone frequency dur : float duration of noise start : float start time for noise dBSPL : float sound pressure of stimulus FMod : float modulation frequency DMod : float modulation depth percent phaseshift : float modulation phase seed : int seed for random number generator Returns ------- array : waveform """ irpts = int(rt * Fs) mxpts = len(t)+1 pin = pipnoise(t, rt, Fs, dBSPL, dur, start, seed) env = (1 + (DMod/100.0) * np.sin((2*np.pi*FMod*t) - np.pi/2 + phaseshift)) # envelope... pin = linearramp(pin, mxpts, irpts) env = linearramp(env, mxpts, irpts) return pin*env
[docs]def linearramp(pin, mxpts, irpts): """ Apply linear ramps to *pin*. Parameters ---------- pin : array input waveform to apply ramp to mxpts : int point in array to start ramp down irpts : int duration of the ramp Returns ------- array : waveform Original (adapted from Manis; makeANF_CF_RI.m):: function [out] = ramp(pin, mxpts, irpts) out = pin; out(1:irpts)=pin(1:irpts).*(0:(irpts-1))/irpts; out((mxpts-irpts):mxpts)=pin((mxpts-irpts):mxpts).*(irpts:-1:0)/irpts; return; end """ out = pin.copy() r = np.linspace(0, 1, irpts) irpts = int(irpts) # print 'irpts: ', irpts # print len(out) out[:irpts] = out[:irpts]*r # print out[mxpts-irpts:mxpts].shape # print r[::-1].shape out[mxpts-irpts-1:mxpts] = out[mxpts-irpts-1:mxpts] * r[::-1] return out
[docs]def pipnoise(t, rt, Fs, dBSPL, pip_dur, pip_start, seed): """ Create a waveform with multiple sine-ramped noise pips. Output is in Pascals. Parameters ---------- t : array array of time values rt : float ramp duration Fs : float sample rate dBSPL : float maximum sound pressure level of pip pip_dur : float duration of pip including ramps pip_start : float list of starting times for multiple pips seed : int random seed Returns ------- array : waveform """ rng = np.random.RandomState(seed) pin = np.zeros(t.size) for start in pip_start: # make pip template pip_pts = int(pip_dur * Fs) + 1 pip = dbspl_to_pa(dBSPL) * rng.randn(pip_pts) # unramped stimulus # add ramp ramp_pts = int(rt * Fs) + 1 ramp = np.sin(np.linspace(0, np.pi/2., ramp_pts))**2 pip[:ramp_pts] *= ramp pip[-ramp_pts:] *= ramp[::-1] ts = int(np.floor(start * Fs)) pin[ts:ts+pip.size] += pip return pin
[docs]def piptone(t, rt, Fs, F0, dBSPL, pip_dur, pip_start): """ Create a waveform with multiple sine-ramped tone pips. Output is in Pascals. Parameters ---------- t : array array of time values rt : float ramp duration Fs : float sample rate F0 : float pip frequency dBSPL : float maximum sound pressure level of pip pip_dur : float duration of pip including ramps pip_start : float list of starting times for multiple pips Returns ------- array : waveform """ # make pip template pip_pts = int(pip_dur * Fs) + 1 pip_t = np.linspace(0, pip_dur, pip_pts) pip = np.sqrt(2) * dbspl_to_pa(dBSPL) * np.sin(2*np.pi*F0*pip_t) # unramped stimulus # add ramp ramp_pts = int(rt * Fs) + 1 ramp = np.sin(np.linspace(0, np.pi/2., ramp_pts))**2 pip[:ramp_pts] *= ramp pip[-ramp_pts:] *= ramp[::-1] # apply template to waveform pin = np.zeros(t.size) ps = pip_start if ~isinstance(ps, list): ps = [ps] for start in pip_start: ts = int(np.floor(start * Fs)) pin[ts:ts+pip.size] += pip return pin
[docs]def clicks(t, Fs, dBSPL, click_duration, click_starts): """ Create a waveform with multiple retangular clicks. Output is in Pascals. Parameters ---------- t : array array of time values Fs : float sample frequency (Hz) click_start : float (seconds) delay to first click in train click_duration : float (seconds) duration of each click click_interval : float (seconds) interval between click starts nclicks : int number of clicks in the click train dspl : float maximum sound pressure level of pip Returns ------- array : waveform """ swave = np.zeros(t.size) amp = dbspl_to_pa(dBSPL) td = int(np.floor(click_duration * Fs)) nclicks = len(click_starts) for n in range(nclicks): t0s = click_starts[n] # time for nth click t0 = int(np.floor(t0s * Fs)) # index if t0+td > t.size: raise ValueError('Clicks: train duration exceeds waveform duration') swave[t0:t0+td] = amp return swave
[docs]def fmsweep(t, start, duration, freqs, ramp, dBSPL): """ Create a waveform for an FM sweep over time. Output is in Pascals. Parameters ---------- t : array time array for waveform start : float (seconds) start time for sweep duration : float (seconds) duration of sweep freqs : array (Hz) Two-element array specifying the start and end frequency of the sweep ramp : str The shape of time course of the sweep (linear, logarithmic) dBSPL : float maximum sound pressure level of sweep Returns ------- array : waveform """ sw = scipy.signal.chirp(t, freqs[0], duration, freqs[1], method=ramp, phi=0, vertex_zero=True) sw = np.sqrt(2) * dbspl_to_pa(dBSPL) * sw return sw