Sound from Scratch

May 2025

Perl's Part in the Game

• Sound propagates through air as pressure changes
• Speakers create pressure changes from variations in voltage
• Sound cards create variations in voltage from data

Our Data Format: LPCM

• LPCM = Linear Pulse Code Modulation
• Data are provided as discrete values (samples)
• Metadata:
  • Number of samples per second, e.g. 44100 (CD) or 48000 (DVD)
  • Binary format of the samples, e.g. 16-bit signed integer
• The .wav file format is basically LPCM plus metadata

Main Ingredients

• Perl with CPAN:
  • Feature::Compat::Class for Corinna syntax
  • The Perl Data Language PDL
  • The GUI toolkit Prima for display and user interfaces
  • Some helpers from the MIDI and Music hierarchies
• SoX - Sound eXchange portable interface to sound cards

?

• Audio::Aoede is the current name of the packages
  • Aoede is the Boeotian muse of voice and song
• sequence is a constructor for PDL objects
  • sequence($n) creates a ndarray with $n elements and values between 0 and $n-1
  • PDL overloads arithmetics and functions to operate on each element of the ndarray
• play is an interface to SoX

?

• short is a PDL data converter
  • Creates a PDL object of "C" shorts (16-bit signed integers)
  • Take care for the value range!
• get_dataref is a PDL method which returns a reference to a Perl scalar containing the binary data
  • No need to pack Perl scalars!

?

• PDL::Graphics::Prima::Simple is a simple (!) way for interactive data visualization
  • line_plot can work with just one PDL object
  • "interactive" not shown in the slides
• slice is a PDL method to access parts of a ndarray
  • Can also be used to change parts of an ndarray

?

The spectrum method uses PDL::FFT (contained in the PDL distribution) to do Fourier transformations

• The method returns a power spectral density (PSD)
• The maximum available frequency of the spectrum is half the sample rate ("Nyquist frequency")
• The resolution of the spectrum in Hz is the sample rate divided by the number of samples (Küpfmüller's uncertainty principle)

Noise...

• ... is sort of a random signal
• ... has colors named after their power spectra:
  • White noise has equal power per frequency interval
  • Pink noise has equal power per halftone
  • There's also brown, blue and violet noise with different power distributions
• So, let's start with the spectrum and calculate the function

...via Aoede

Envelope Parameters

• ADSR = Attack, Decay, Sustain, Release
  • Attack: How fast the sound reaches peak volume
  • Decay: How fast the sound fades
  • Sustain: What volume wil be sustained (a level, not a time)
  • Release: How fast the sound dies down

More Features...

Effects

• Vibrato - periodically change the frequency
• Tremolo - periodically change the amplitude
• Non-harmonic overtones
• Non-periodic functions

Many Distractions...

• "Interactive" Sound
  • Adds quite some complexity
• "Can you do frequency analysis in real time?"
  • Short answer: Yes.
• "That needs some drums!"
  • Can be done, but is a bit tedious to do from scratch.
• "What about MIDI?"
  • There's some hacks, with help of CPAN modules.

Interactive Sound

• Play sound while a key is pressed
  ...and immediately stop when it is released
• Pause and resume playing
• Needs an event loop with a timer
  • Hence the "source" interface:
    $source->next_samples($n,$first)