sk de en
Main page > Hobby Projects and Research > Choir-Sound Modelling

Choir-Sound Modelling

© Dipl.-Ing. Radovan Konečný - radkon, 2013-2014

About the project

This project was created in order to make the choir sound only with help of minimal means, i.e. without recordings of an existing choir.

In the first analysis phase, I acquired the spectral composition of sound for individual tones. In the case of a human voice, it is very important to make it separately for each tone. I sang and recorded every tone. A few (e.g. 4) signal periods were then selected from the recording and a single period was calculated as the average of these selected periods. From this single (average) period, the spectral composition (i.e. the amplitudes of the individual harmonic components) was calculated by the Fourier transform. The result of the analysis is just a few hundreds of numeric values for each tone.

In the second synthesis phase, it was necessary to create a sound that should satisfy several properties:

  • a character similar to a singing choir with several singers,
  • perfectly cyclical (i.e. when playing repeatedly there should be no audible seam-point),
  • one loop of cyclical sound should not be long,
  • the ability to create a nice stereo sound.

To synthesize sound that should meet these qualities, I used my recent project Sound modelling - NoiseSaw, where I added the ability to specify the spectral composition of the sound for each particular tone.

In this way, I have made a resynthesis of my own voice, but it sounds rather like a choir. Tones do not even have to be sung intonation accurately - the used method of analysis and synthesis will always guarantee perfect tuning. The limitation was only the height range of my voice (C2 to G#4), but it is no problem, if the aim was to resynthesize a particular voice. This means that the tones I cannot sing cannot be played.

- sample - the resulting synthesized sound from the spectrum of my voice

- resynthesis procedure - first sung tone, followed by the tone of the selected period and finally synthesized sound from this period


In this way, it is easy to make the voice resynthesis of anyone. Alternatively, it is possible to make a choir, mixed of multiple voices or of multiple types of voice (male, female, ...).

Usage in KORG M1

If it is possible to create sound loops with a length of about 2 seconds, it is possible to insert such sound into my remade KORG M1 synthesizer. The used synthesis can create loops for any sample rate, which is no problem for the 31.25 kHz sample rate in the KORG M1.

- sound comparison - first original (legendary) sound from KORG M1, then resynthesized my voice inserted into the remade KORG M1

Usage in Sampler

Longer loops could also be produced for the Sampler (5 seconds look to be sufficient) and stereo sound could be produced (the synthesis directly allows this). The highest mentioned sample is just using sound loops inserted into my Sampler.

Usage in the Struck instrument simulator

Since my Struck instrument simulator can also create sounds of this character, it was also possible to produce a model for the Struck instrument simulator. This creates real-time audio.

- sample

The model file is on the project page. Just for interest - the model file is only 33 kB large.

Copyright © 2009-2018 Dipl.-Ing. Radovan Konečný - radkon. All rights reserved.