I did something I should have done awhile ago, and that’s invest into a SoundCloud account. I recently rendered all the Csound Blog examples and uploaded them. So now it’ll be easier for many of you to listen to clips without having to render them yourselves.

The Csound Blog Set List at SoundCloud

Now that the Slipmat dev code is up at github, I can write smaller examples rather than cram everything into one large script. The example test_tone.py is written specifically to demonstrate the simplest possible script that generates audio. Here’s the code:

import slipmat

s = slipmat.ScoreEvents()
s.event(0, 4, slipmat.Sine())
slipmat.ScoreEventsToWave(s, "./test_tone.wav")

This produces a 4-second 440Hz sine tone, and writes it to a wave file.

Here’s the line-by-line run down. First, the slipmat module is imported, essentially turning Python into a synthesizer. Then a score object is created from class slipmat.ScoreEvents(); this is akin to a Csound score file, as it is responsible for scheduling events for unit generators and instruments. A slipmat.Sine() object is instantiated then scheduled to play for 4-seconds using the default frequency of 440Hz. Then slipmat.ScoreEventsToWave() parses the score and renders the audio, writing the results to the file “./test_tone.wav”.

This is how it looks today. As Slipmat changes and evolves, so will this example.

If RSS isn’t your cup of tea, you can alternatively follow _slipmat @ twitter. All new blog posts, along with git commits (in digest form) are set to auto-aggregate to there. I’ll also post links and information related to slipmat, computer music and audio synthesis from time to time. And you’ll be able to send tweets my way as well.

By utilizing a Python list as a stack, it is possible to move the stylus to a specific time in the score, schedule events relative to the stylus, and then pop it back into its previous position. This is inspired by Processing functions pushMatrix() and popMatrix().

I’ll explain what I’m talking about through example. The following is an excerpt from the pushpop.py example in the Slipmat github repo:

# Measure 1
s.event(0, 0.25, Sine(1, cpspch(7.00)))
s.event(1, 0.25, Sine(1, cpspch(7.03)))
s.event(2, 0.25, Sine(1, cpspch(7.07)))
s.event(3, 0.25, Sine(1, cpspch(7.10)))

# Measure 2
s.time.append(4)
s.event(0, 0.25, Sine(1, cpspch(8.00)))
s.event(1, 0.25, Sine(1, cpspch(8.03)))
s.event(2, 0.25, Sine(1, cpspch(8.07)))
s.event(3, 0.25, Sine(1, cpspch(8.10)))
s.time.pop()

In measure 1, four Sine() events are created at times 0, 1, 2 and 3, specified in the first parameter of s.event().

In measure 2, before any calls to s.event() are made, the value 4 is append to a list object called “time,” part of class ScoreEvents. Until “time” is popped or appended, the value 4 is added to every start time in method event(). Measure 2 generates 4 events, created at times 4, 5, 6 and 7.

If these two measures were explicitly written, it would look like this:

s.event(0, 0.25, Sine(1, cpspch(7.00)))
s.event(1, 0.25, Sine(1, cpspch(7.03)))
s.event(2, 0.25, Sine(1, cpspch(7.07)))
s.event(3, 0.25, Sine(1, cpspch(7.10)))
s.event(4, 0.25, Sine(1, cpspch(8.00)))
s.event(5, 0.25, Sine(1, cpspch(8.03)))
s.event(6, 0.25, Sine(1, cpspch(8.07)))
s.event(7, 0.25, Sine(1, cpspch(8.10)))

The current position of the stylus is the combined sum of all values stored in list “time.” For example:

s.time.append(16)
s.time.append(30)
s.time.append(11)
s.event(7, 1, foo())
s.time.pop()
s.event(2, 1, bar())

The foo() event is scheduled at beat 64. That’s (16 + 30 + 11) for “time” plus an additional 7 for the first parameter of event(). After “time” is popped, bar() plays at beat 48; (16 + 30) for “time” plus 2 for the relative start of the event.

For the record, I think the s.event() code is a bit ugly, which I’m already addressing behind the scenes. If you’ve been keeping up with this blog, you might already have a good idea as to where I’m going with this.

Slipmat code has a new home at github. I have few things to get in order before I push the code to the new repository. Things should be up sometime week.

Just to be absolutely clear, the blog is staying here at noisepages.

I had three goals in mind in getting the first working Python Slipmat prototype, Lead-In, up and running:

• Use pure Python, with no external modules.
• Getting Python to behave like a real audio langauge.
• Mold the syntax to feel like a real audio language.

I would say it’s a success on all three fronts, with plenty of room to grow.

However, there are some major issues. The biggest being that it’s slow; Amazingly slow. This was expected, as Python itself is an interpreted language, and not a DSP power house like C/C++. On my modern iMac, slipmat_lead-in.py took 1 minute and 19.681 seconds to render. I bet Csound would do a comparable job, in 1996.

So one of my first goals of the upcoming redesign is to make it just fast enough until a proper C/C++ engine can be written. I’m looking into NumPy/SciPy.

I’m also looking into Audiolab, “a python package to make noise with numpy arrays,” written David Cournapeau. This is, more or less, a Python wrapper module for Erik de Castro Lopo’s excellent libsndfile; I was already planning on using libsndfile, anyhow.

Slipmat, now with audio.

Python Script: here