ISMIR and DAFx Tutorials 2012

h1. Ismir2012: Software Session Plan

h2. Length and plan

* 90 min (75 + 15m interval)

h2. Software Requirements

* Python 

* Numpy

* Matplotlib

* Nose and Nosetests

* ipython (desirable, not mandatory)

h2. Repository

https://code.soundsoftware.ac.uk/hg/python-tutorial-skeleton

h2. Script

* Clone/Check if repository was successfully cloned

* Explore folder and files 

* Launch python shell (ipyhon)

* import audiofile.py and plot samples

* Explain what the test files are and introduction to nose 

* Create estimate_tempo.py (main programme)

<pre>

import tempo_estimator as est

import sys

filename = sys.argv[1]

tempo = est.estimate_tempo_of_file(filename)

</pre> 

* Easy Mercurial: 

** ignore pyc and py~

** commit estimate_tempo.py

** commit .hgignore

* Create estimate_tempo.py

* Create test file test_estimate_tempo.py

<pre>

import estimate_tempo as est

def test_tempo_120bpm():

    tempo = est.estimate_tempo_of_file("test files/120bpm.wav")

    # we consider the test to be passed if error is less than 1/4 of bpm

    error = abs(tempo - 120) < 0.25

    assert 

</pre>

* Commit both files

* Start writing the code of tempo_estimator.py

<pre>

</pre>

*** *Slide*: how to detect tempo?

*** *Slide*: how our estimator will work

* Write the onset detection function and autocorrelation function calls in tempo_estimator.py

<pre></pre>

* Commit

* Python Shell: 

** import the audiofile module

** load the audiofile

<pre>

import audiofile as af

(samples, samplerate) = af.read_all_read_all_mono_samples_from_file('test files/beatbox.wav')

</pre>

** split the file into non overlapping blocks and calculate the RMS of each

** Hot to calculate the RMS?

** Create signal_processing.py and test_signal_processing.py

*** *Slide*: how our estimator will work

** ipython: write rms function 

<pre>

import numpy as np 

def rms(array):

    return np.sqrt(np.mean(samples**2))

</pre>

** write RMS tests 

** paste RMS function to signal_processing.py

** nosetests signal_processing.py

* in ipython again

** how many frames (non overlapping) does the signal have?

<pre>np.ceil(samples/512)</pre>

*** don't forget to mention tghat ceil returns a float

** what are the contents of the first frame?

<pre>samples[0:512]</pre>

** what are the contents of the second frame?

<pre>samples[512:1024]</pre>

** Generalising....

* signal_processing.py

** def get_frame_count(n_samples, increment)

** def get_frame(samples, increment, n)

** write tests for both functions

** ipython: protoype both functions, write them, test them

* implement detection function on tempo_estimator.py

* implement tests

* Autocorrelation+ testing

* plot waveform and detection function

** time scales are wrong

* create plots.py just as a simple plotting script

* in tempo_estimator.py create bpm_to_lag and lag_to_bpm functions

* add to tempo estimator definition bpm_min and bpm_max arguments