In "Your brain hallucinates your conscious reality," Anil Seth's demo of the brain being able to understand a completely distorted audio clip was amazing.

(Just watch 60 seconds from this point rather than the full 17 minute lecture.)

If I want to make my own audio clip demo like that, how can I distort a recording of my voice in the same way that Anil did?

(I'm having trouble finding the best settings in Audacity.)

  • I'm not surprised we can understand the synthesized signal. I believe that modulations is what carries all the information in a signal. So even if we were to create a new signal with the same basic modulations, we are still transmitting the same information. – Schizomorph Jun 14 at 15:19

I wouldn't call this a distortion as it is a 'new', synthesized sound based on the original sound. A distortion would be the original sound, shaped (see waveshapers)

I believe this is a sine wave that is modulated based on the characteristics of the original phrase. It sounds like the pitch follows the original pitch. The envelope (the way the amplitude varies through time) follows the original envelope and there was an attempt to do the same with timbre by something that sounds like FM synthesis to my ears.

I can't give you an exact 'recipe' for this patch but I would start with a digital modular synthesizer, Reaktor or Max MSP since this kind of tools allow the flexibility to build a patch like this.

The easiest part is the envelope - use an envelope follower and route it to your final amplitude, usually a VCA.

Pitch is quite tricky and I can't tell you how to do it but in principle, you need to use FFT to analyze the incoming signal, find the highest peak in the spectrum diagram and plot how it moves against time. This curve needs to control the pitch of a sine wave oscillator.

Timbre is also tricky. If I'm getting this right, they are using the amplitude of the incoming signal (modulator) to control the amplitude of fixed frequency sine wave oscillator which in turn modulates the frequency of our first oscillator.

I found a really impressive free website that is almost ¹ good enough to use for this purpose: https://voicechanger.io/

The one I used is called "alien2" and is currently shown as the 8th option from the top left.

P.S. Behind the scenes (in the code of "alien2"), these seem to be the relevant settings (although I don't know what they mean):

oscillator.frequency.value = 20;
oscillator.type = 'sine';
oscillatorGain.gain.value = 0.01;
delay.delayTime.value = 0.05;

"robot3" is another decent option.

¹ But I'd prefer to distort the audio so much that an unsuspecting listener wouldn't even assume (on the first time listening to it) that words were being spoken, and then after hearing the original audio, she would feel like it obviously contained words (and even which words were obvious).

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Not the answer you're looking for? Browse other questions tagged or ask your own question.