Thursday, 6 November 2014

Producing formant noise words to order

Nothing spectrogramFollowing on from my last post, I had a thought about formant noise, a phenomenon which must be eliminated as a possible cause when analyzing EVP recordings. My own experiments suggest that rhythm is more important than formant-like frequency peaks in producing words. I don't think formant noise works exactly like ordinary human speech. With formant noise, I think the brain hears an apparent bit of speech and tries to fit a word onto it from memory. This has parallels with visual substitution and explains why interpretations can drift - sometimes you hear one phrase, at other times another, from the same sound.

In formant noise, I think the frequency peaks are primarily important in switching the brain into speech mode. I don't think it matters too much which peaks are present, so long as there are some. I believe it is the rhythm that produces the actual words. So, I think it should be possible to produce specific words or phrases to order, by using just the right rhythm. I decide to try this idea.

My aim was to produce a formant noise clip that said the word 'nothing'. It's a simple word with an easily identifiable rhythm to it. I used various mechanical methods for producing sounds with no real voices involved. I have identified, in the past, certain mechanical sounds (that occur normally widely) that tend to produce better formant noise than others, so naturally I used those.

I recorded lots of attempts to produce the word 'nothing' but only found a couple which, at first hearing, sounded promising. Here is the first one : sample 1. Now here's the strange thing. It sounded like the word 'nothing' when played on the original recording with all the other samples. But when I isolated it sounded quite different. My interpretation is at the end of this post, so try listening to the sample first, to see what you think, before seeing what I heard. The spectrogram (above right) shows frequency versus amplitude for sample 1. You can see quite clear multiple frequency peaks occurring simultaneously, typical of both real speech formants and formant noise.

The same thing happened with the second promising attempt: sample 2. It only sounded like 'nothing' before it was isolated from the other attempts. On its own, it sounded quite different (see below for my interpretation). But then something even weirder happened. On listening this sample again, an hour or so later, I could no longer hear at as my original interpretation, even that one sounded perfectly stable and strong earlier. Instead it sounded like another word entirely (see below). But now, later again, it has reverted to my first - non-nothing - interpretation.

The change from the word 'nothing' to something quite different, when isolated, shows the importance of context. Clearly the sounds preceding the samples modified how they were heard. The drift in interpretation of sample 2 is typical of formant noise.

So, I still haven't produced the word 'nothing' yet. I suspect I am not getting the rhythm quite right. I still think it's possible though.

My interpretations: Sample 1 has always sounded like 'natural rhythm' to me. This might be because the word rhythm was floating around my head during this experiment! Sample 2 originally sounded like 'I'm fine' but then changed to 'reverie'. One interpretation or the other persists until I try again later.

No comments:

Post a Comment