posted 20 July 2009 08:47            

There is a unix command (OSX, Linux); md5, whereby providing the command with a string generates a unique hexadecimal number for the string.

e.g. andrew# md5 -s kyma
MD5 ("kyma") = 31a27630697a5ab56081b40035733a85

Perhaps if this number were then split into 16 parts, and the audio signal split into 16 bands. Each part of the original md5 number could be used to alter the source signal (using some spectral distortion on each band?) in order to make the resulting (sum of the 16 bands) signal unintelligible.

The listener of the audio would apply the reverse algorithm, having been informed of the string for use with md5 to provide the unlock key.

Would this be feasible? Or would it not be possible to undo the effects of a scrambled audio signal?

posted 20 July 2009 10:10            

For encryption you'd want to use use something like AES. This can be symmetrical or asymmetrical encryption. asymmetrical would be the most secure with each party having a different private key for decryption and a public key that is shared.

AES 256 is pretty darn secure if used properly.

Thing is, in an internet context, true streaming audio is many times done on a lossy connection by udp rather than tcp so not every packet is guaranteed to arrive at destination. This complicates the matter in an encryption scenario because you need a lossless transmission for unencryption unless the streaming protocol supports encryption natively at the packet level.

posted 21 July 2009 14:16            

Don't know if this is the sort of thing you meant but you can do odd things with the single side band ring modulator. That is the true one using the Hilbert transform not the other one in the prototypes.

With this you can pitch shift by half sample rate. This will make spectral envelope turn upside down in pitch and the sound will be intelligible.

If you do the same thing again you get your original signal back again.

Not much of an encoding system when there is only one way of screwing it up. If instead you keep both in phase and 90 degs out of phase, you have to keep both signals but you can pitch shift it by any frequency and you can get the original back.

The odd thing is that the pitch shift moves the frequencies around in a big circle held between DC and Half sample rate. It has both negative and positive frequencies as long as you keep both in phase and 90 deg out of phase signals.

With careful selection of pitch shifts you can add the results together in a way that can be decoded. For example shift by quarter sample rate and half sample rate then by shifting the mix by another quarter sample rate and mixing with it self should make one copy cancel and the other copy add giving back the original signal.

Don't know if this is the sort of thing you are looking for?

posted 22 July 2009 07:41            

thanks for your pointers on methods which would make this possible.

but yes I am making the distinction between encrypting the data stream (e.g. a secure TCP/IP connection) carrying the audio samples and the audio samples contained within that data stream which perhaps could be encrypted/decrypted within a kyma sound.

This could negate the side-effects of a UDP connection; the encryption/decryption of the audio samples would be independent of the network connection - not all packets would need to arrive - the kyma sound would just be working on audio samples, if some went missing, it would just sound like audio breaking up, wouldn't it?