Inverting the audio spectrum

Special Interest Groups Audio Processing
1

I have some very old audio files with Soviet cosmonauts talking in inverted spectrum voice. I used to have an analog device that did the unscrambling but it no longer exists. I understand that by inverting the sign of every second sample the spectrum will be inverted. Has anyone ever developed an add-on to Audacity to do this? If not, are there any similar add-ons whose code can be partly re-used to do this neat trick?

BTW the account of my old analog inversion device can be found here:

http://www.svengrahn.pp.se/trackind/scramble/scramble.htm

Sven

2

I don’t speak Russian so I don’t know if this is intelligible or not.

I applied the following code using the Nyquist Prompt effect (http://manual.audacityteam.org/o/man/nyquist_prompt.html)

(lowpass8 (mult 2 (highpass8 s 300) (hzosc 3763)) 3100)
3

OK, so you implemenetd my hardware solution in code - clever. It sounds pretty good. One can fiddle around with filter edges and LO fx, but it should be good. I will try te other files I have got. Thanks a million. Have to read up more on the lisp syntax…

Sven

4

It would be worth playing with the carrier frequency. I seem to be getting clearer speech with a slightly lower carrier frequency (around 3700 Hz).

There’s some reading material here :wink: https://forum.audacityteam.org/t/manuals-and-reference-material/32817/1

5

I understand the code line except the number “2” after mult. Please explain!

6

“Multiply by 2”

MULT can act on numbers or sounds.
When the there is (MULT number sound), each sample value of sound is multiplied by number.
In other words, the sound is amplified. In this case, by +6.0206 dB (x 2 on a linear scale).

http://www.cs.cmu.edu/~rbd/doc/nyquist/part8.html#index589

7

Yes, I understand “2” now. I have tried various settings by modifying your code snippet to make it better. For example 3740 Hz is a Little bit better in my view. 3700 Hz sounds a Little “Donald Duckish”. I need to experiment with resampling the audio and filtering it before trying to invert. It is sort of unneccesary to have spectral components above 3.5 kHz or so, I guess. So maybe I should resample at 6000 or 8000 Hz? Many thanks again.