It became evident from the previous experiment that used actual words that it was not difficult to arrange them to make meaningful sentences. This, coupled with the rather non-random nature of the RNG when doing the 900 ms experiment may have led to false results.

Accordingly, the experiment was repeated using nonsense words (as far as common usage was concerned).

The results were as follows.

There were typical sounds at 450 but there were just no words in English.

Strangely, at 900, meaningful phrases re-appeared.

'Low Pal Got Her'

'Let my hook reason him'

(Remember the Hook Operator in Barbra O'Brien's 'Operators and Things').

Once again at 900 there was a sequence repetition, probably 1, 2, 1, 2.

It may be that the RNG has trouble with only 13 samples being available.

I am asking it to sample at a rate that is close to the repetition rate of the individual words - 1000/900 compared with 1000/385(nom). Possibly some effect rather like aliasing is happening to interfere with the randomity.

The zero result at 450 is not unexpected. Here, the rate is so close to the nonsense word rate that for the result to be also nonsense is fair enough.

The faster rates are interesting. Because of the improvement I decided to extend the sequence down to 45 and then 10 ms.

The surprise is 45 - this is showing up as the relative best, and the three-point improvement curve adds some credibility to this finding.

The graph below indicates how the conversion efficiency varies with slice period, periods displayed on a logarithmic scale.

Remember these are just quick informal experiments.

I have to leave this now to get on with some real world work.

Alec

Tus

roong

Ter

dal

Mowk

rees

bim

Laif

Mive

lo

pow

gaw

sal