Skip to content

Latest commit

 

History

History
233 lines (193 loc) · 15.2 KB

README.md

File metadata and controls

233 lines (193 loc) · 15.2 KB

Package PyPI Documentation Status Tutorial Maintenance PyPI pyversions PyPI license DOI

slab: easy manipulation of sounds and psychoacoustic experiments in Python

Important

Version 1.5 changes the Filter and HRTF interfaces very slightly. Existing code should be updated in the following way:

  1. HRTF.interpolate now returns a Filter instead of an HRTF object with a single source, so call hrir.apply() instead of hrtf[0].apply().
  2. The fir attribute of the Filter class now takes a string to indicate the type of filter ('FIR', 'IR', 'TF') instead of a bool. Use fir='TF' where you previously used fir=False, and fir='FIR' or 'IR' where you used fir=True, depending on whether the filter should be applied using scipy.signal.filtfilt (intended for highpass filters and the like) or scipy.signal.fftfilt (intended for HRTFs and room responses).

Slab ('es-lab', or sound laboratory) is an open source project and Python package that makes working with sounds and running psychoacoustic experiments simple, efficient, and fun! For instance, it takes just eight lines of code to run a pure tone audiogram using an adaptive staircase:

import slab
stimulus = slab.Sound.tone(frequency=500, duration=0.5) # make a 0.5 sec pure tone of 500 Hz
stairs = slab.Staircase(start_val=50, n_reversals=10) # set up the adaptive staircase
for level in stairs: # the staircase object returns a value between 0 and 50 dB for each trial
    stimulus.level = level
    stairs.present_tone_trial(stimulus) # plays the tone and records a keypress (1 for 'heard', 2 for 'not heard')
print(stairs.threshold()) # print threshold when done

Why slab?

The package aims to lower the entrance barrier for working with sounds in Python and provide easy access to typical operations in psychoacoustics, specifically for students and researchers in the life sciences. The typical BSc or MSc student entering our lab has limited programming and signal processing training and is unable to implement a psychoacoustic experiment from scratch within the time limit of a BSc or MSc thesis. Slab solves this issue by providing easy-to-use building blocks for such experiments. The implementation is well documented and sufficiently simple for curious students to understand. All functions provide sensible defaults and will many cases 'just work' without arguments (vowel = slab.Sound.vowel() gives you a 1-second synthetic vowel 'a', vowel.spectrogram() plots the spectrogram). This turned out to be useful for teaching and demonstrations. Many students in our lab have now used the package to implement their final projects and exit the lab as proficient Python programmers.

Features

Slab represents sounds as Numpy arrays and provides classes and methods to perform typical sound manipulation tasks and psychoacoustic procedures. The main classes are:

Signal: Provides a generic signal object with properties duration, number of samples, sample times, number of channels. Keeps the data in a 'data' property and implements slicing, arithmetic operations, and conversion between sample points and time points.

sig = slab.Sound.pinknoise(n_channels=2) # make a 2-channel pink noise
sig.duration
# 1.0
sig.n_samples
# 8000
sig2 = sig.resample(samplerate=4000) # resample to 4 kHz
env = sig2.envelope() # returns a new signal containing the lowpass Hilbert envelopes of both channels
sig.delay(duration=0.0006, channel=0) # delay the first channel by 0.6 ms

Sound: Inherits from Signal and provides methods for generating, manipulating, displaying, and analysing sound stimuli. Can compute descriptive sound features and apply manipulations to all sounds in a folder.1

vowel = slab.Sound.vowel(vowel='a', duration=.5) # make a 0.5-second synthetic vowel sound
vowel.play() # play the sound
vowel = vowel.ramp() # apply default raised-cosine onset and offset ramps
vowel = vowel.filter(kind='bp', frequency=[50, 3000]) # apply bandpass filter between 50 and 3000 Hz
vowel.spectrogram() # plot the spectrogram
vowel.spectrum(low_cutoff=100, high_cutoff=4000, log_power=True) # plot a band-limited spectrum
vowel.waveform(start=0, end=.1) # plot the waveform
vowel.write('vowel.wav') # save the sound to a WAV file
vocoded_vowel = vowel.vocode() # run a vocoding algorithm
vocoded_vowel.play() # play the vocoded sound
vowel.spectral_feature(feature='centroid') # compute the spectral centroid of the sound in Hz
# [1016.811]

Binaural: Inherits from Sound and provides methods for generating and manipulating binaural sounds, including advanced interaural time and intensity manipulation. Binaural sounds have left and a right channel properties.

sig = slab.Binaural.pinknoise()
sig = sig.pulse() # make a 2-channel pulsed pink noise
sig.n_channels
# 2
right_lateralized = sig.itd(duration=600e-6) # add an interaural time difference of 600 µsec, right channel leading
# apply a linearly increasing or decreasing interaural time difference.
# This is achieved by sinc interpolation of one channel with a dynamic delay:
moving = sig.itd_ramp(from_itd=-0.001, to_itd=0.01)
hrtf = slab.HRTF.kemar() # using the default head-related transfer function
level_spectrum = slab.Binaural.make_interaural_level_spectrum(hrtf) # compute frequency-band-specific ILDs from KEMAR
lateralized = sig.at_azimuth(azimuth=-45, ils=level_spectrum) # add frequency-dependent ITD and ILD corresponding to a sound at 45 deg
external = lateralized.externalize() # add an under-sampled HRTF filter that results in the percept of an external source
# (i.e. outside of the head), defaults to the KEMAR HRTF recordings, but any HRTF can be supplied

Filter: Inherits from Signal and provides methods for generating, measuring, and manipulating FIR and FFT filters, filter banks, and transfer functions.

sig = slab.Sound.whitenoise()
filt = slab.Filter.band(frequency=2000, kind='hp') # make a highpass filter
filt.tf() # plot the transfer function
sig_filt = filt.apply(sig) # apply it to a sound
# applying a whole filterbank is equally easy:
fbank = slab.Filter.cos_filterbank(length=sig.n_samples, bandwidth=1/10, low_cutoff=100) # make a cosine filter bank
fbank.tf() # plot the transfer function of all filters in the bank
subbands = fbank.apply(sig) # make a multi-channel sound containing the passbands of the filters in the filter bank
subbands.spectrum(low_cutoff=90) # each band is limited by the corresponding fbank filter
# the subbands could now be manipulated and then combined with the collapse_subbands method
fbank.filter_bank_center_freqs() # return the centre frequencies of the filters in the filter bank
fbank = slab.Filter.equalizing_filterbank(reference, measured) # generate inverse filters to minimize the difference
# between measured signals and a reference sound. Used to equalize loudspeakers, microphones, or speaker arrays.
# measured is typically a recorded signal (potentially multi-channel), and reference for instance a flat white noise.
fbank.save('equalizing_filters.npy') # saves the filter bank as .npy file.

HRTF: Inherits from Filter, reads .sofa format HRTFs and provides methods for manipulating, plotting, and applying head-related transfer functions.

hrtf = slab.HRTF.kemar() # load in-built KEMAR HRTF
print(hrtf) # print information
# <class 'hrtf.HRTF'> sources 710, elevations 14, samples 710, samplerate 44100.0
sourceidx = hrtf.cone_sources(20) # select sources on a cone of confusion at 20 deg from midline
hrtf.plot_sources(sourceidx) # plot the sources in 3D, highlighting the selected sources
hrtf.plot_tf(sourceidx,ear='left') # plot transfer functions of selected sources in a waterfall plot
dtf = hrtf.diffuse_field_equalization() # apply diffuse field equalization to remove non-spatial components of the HRTF

Room: Easy simulation of echoes and reverberation.

room = slab.Room(size=[4,6,3], listener=[2,3,1.8], source=[0,0,1])  # create an echo list for a room size, listener ans source position
hrir = room.hrir()  # compute the room impulse response (as an slab.Filter)
sound = slab.Sound.vowel(duration=0.3, samplerate=hrir.samplerate)  # create an example sound to add the reverb to
echos = hrir.apply(sound)  # apply the room impolse response

Psychoacoustics: A collection of classes for working trial sequences, adaptive staircases, forced-choice procedures, stimulus presentation and response recording from the keyboard and USB button boxes, handling of precomputed stimulus lists, results files, and experiment configuration files.

# set up an 1up-2down adaptive weighted staircase with dynamic step sizes:
stairs = slab.Staircase(start_val=30, max_val=40, n_up=1, n_down=2,
                            step_sizes=[3, 1], step_up_factor=1.5)
for trial in stairs: # draw a value from the staircase; the loop terminates with the staircase
    response = stairs.simulate_response(25) # simulate a response from a participant using a psychometric function
    print(f'trial # {stairs.this_trial_n}: intensity {trial}, response {response}')
    stairs.add_response(response) # logs the response and advances the staircase
    stairs.plot() # updates a plot of the staircase in each trial to keep an eye on the performance of the listener
stairs.reversal_intensities # returns a list of stimulus values at the reversal points of the staircase
stairs.threshold() # computes and returns the final threshold
stairs.save_json('stairs.json') # the staircase object can be saved as a human readable json file

# for non-adaptive experiments and all other cases where you need a controlled sequence of stimulus values:
trials = slab.Trialsequence(conditions=5, n_reps=2) # sequence of 5 conditions, repeated twice, without direct repetitions
trials = slab.Trialsequence(conditions=['red', 'green', 'blue'], kind='infinite') # infinite sequence of color names
trials = slab.Trialsequence(conditions=3, n_reps=20, deviant_freq=0.12) # stimulus sequence for an oddball design
trials.transitions() # return the array of transition probabilities between all combinations of conditions.
trials.condition_probabilities() # return a list of frequencies of conditions
for trial in trials: # use the trials object in a loop to go through the trials
    print(trial) # here you would generate or select a stimulus according to the condition
    trials.present_afc_trial(target, distractor, isi=0.2) # present a 2-alternative forced-choice trial and record the response

stims = slab.Precomputed(lambda: slab.Sound.pinknoise(), n=10) # make 10 instances of noise as one Sound-like object
stims = slab.Precomputed([stim1, stim2, stim3, stim4, stim5]) # or use a list of sound objects, or a list comprehension
stims.play() # play a random instance
stims.play() # play another one, guaranteed to be different from the previous one
stims.sequence # the sequence of instances played so far
stims.write('stims.zip') # save the sounds as zip file of wavs
stims = slab.Precomputed.read('stims.zip') # reloads the file into a Precomputed object

1) The basic functionality of the Signal class and some of the sound generation methods in the Sound class were based on the brian.hears Sound class (now brian2hears, an auditory modelling package).

Installation

Install the current stable release from the python package index with pip: pip install slab

Other requirements

On Linux, there is only one requirement outside of Python: you may need to install libsndfile using your distribution’s package manager, for instance:

sudo apt-get install libsndfile1

On Macs with M1 processors, the SoundCard module that slab uses to play and record sounds is currently not working. You can workaround this issue by uninstalling SoundCard:

pip uninstall soundcard

Slab will fall back to afplay to play sounds. Recording sounds directly from slab is not possible in this case.

Other optional requirements can be installed by telling pip which extras you want:

pip install slab[name_of_extra]

The options for name_of_extra are:

  • windows: if you are running Windows - this will install windows-curses for you, which is required for getting button presses in the psychoacoustics classes,
  • hrtf: if you want to use spatial stimuli with the Binaural and HRTF classes,
  • testing: (for developers) if you want to run the unit tests for slab, and
  • docs: (for developers) if you want to build the documentation locally.

You can combine these options: pip install slab[windows, hrtf] if you are on Windows and use spatial sounds.

Detailed installation instructions can be found here.

You can also get the latest development version directly from GitHub (if you have git) by running: pip install git+https://github.com/DrMarc/slab.git

The releases use semantic versioning: major.minor.patch, where major increments for changes that break backwards compatibility, minor increments for added functionality, and patch increments for internal bug fixes. slab.__version__ prints the installed version.

Documentation

Read the tutorial-style documentation on ReadTheDocs. For an interactive tutorial without installing anything, try the Colab notebook: Open tutorial in Colab

Citing slab

Schönwiesner et al., (2021). s(ound)lab: An easy to learn Python package for designing and running psychoacoustic experiments. Journal of Open Source Software, 6(62), 3284, https://doi.org/10.21105/joss.03284

@article{Schönwiesner2021,
  doi = {10.21105/joss.03284},
  url = {https://doi.org/10.21105/joss.03284},
  year = {2021},
  publisher = {The Open Journal},
  volume = {6},
  number = {62},
  pages = {3284},
  author = {Marc Schönwiesner and Ole Bialas},
  title = {s(ound)lab: An easy to learn Python package for designing and running psychoacoustic experiments.},
  journal = {Journal of Open Source Software}
}

Contributing to this project

Anyone and everyone is welcome to contribute. Please take a moment to review the guidelines for contributing.

License

The project is licensed under the MIT license.