An Introduction to Sine-Wave Speech (2007)

http://www.haskins.yale.edu/featured/sws/sws.html is a page the Haskins Lab folks have put up about this; unfortunately at some point they switched it to use Flash or something, so now it's unusable. The archived version at https://web.archive.org/web/20060828113107/www.haskins.yale.... directly links to AIFF files; by telling Firefox to open them with /usr/bin/play or /usr/bin/vlc I can hear them.

The original experiments were done in the 1970s, I think, and the papers on the page are I think from 1981–1994. It would have been interesting to see how things would have developed if this formant-centric viewpoint had become central to speech codec research in the 80s and 90s. In a sense LPC codecs like GSM RTE-LTP are formant-centric, since the linear prediction coefficients in some sense encode the formant structure of the speech, but (and I'm getting out my depth here) as I understand it the impulse response of the linear prediction filter is pretty broadband.

We really need a better medium than the current WWW for this kind of archival, because the researchers are going to retire or die, and then they will probably stop updating their pages, which will then apparently become inaccessible with modern browsers.

I listened to only the "degraded" sample, and I could understand it on the second listen. English is not my native language and I have pretty severe hearing loss. I suspect the latter makes it easier for me in some situations to decode language with a lot of noise - I have a lot of practice guessing, I suppose. It would be interesting seeing some research on this.

What I find interesting is that I only need priming on one, then I can understand all of them once I realize I'm supposed to be listening for speech that's behind some kind of high-pitched distortion.

It reminds me somewhat of when I'm in a bilingual conversation and somebody unexpectedly switches to the other language for just a quick word or two when you don't expect it. Your brain tries to interpret the sounds in the first language and they're just gibberish, even though you speak the second language. Then if the speaker continues in the second language quickly, you "retroactively" re-interpret what you just heard and understand it.

After hearing the first two examples with clear speech I could identify most of the last three without, took a few listens before the last one clicked.

I saw the other submission yesterday and thought it was very cool. Today it's more meh.

The priming didn't hold for me from yesterday. I clicked the first of the four extra examples and didn't understand anything.

After listening to the clear version and trying again I heard it.

But again priming did not hold for me (unlike what's described on the page): each of the other three examples were unintelligible to me (not even recognizable as speech sounds). Again, after listening to the clear version, it became easy, but only for the corresponding sine-wave encoding.

I still like that idea, if only for nostalgic reasons. We did some formant analysis at university, up to trying to "read" a sonagram, i.e. deducing fron the sonagram as shown on the page what was being said.

Random data point but as a non-native speaker I could understand about 80% of the words on the first go. Listening to the clear audio didn't really make me "hear" the correct word on the 2nd try. It just made me realize that "yes, that word that I know is correct here, might indeed sound like this" IDK how to explain it. Take for example, the "The camel was kept in a cage at the zoo" sentence. I heard "The owl was kept in a cage at the zoo" on the first go and after listening to the clear version, I still hear it's an "owl". I now know that "camel" is also reasonable there and makes sense, but I still hear "owl". Am I wired incorrectly? :)

DIBS! I called it! Dibs on this for robot voices! H.E.L.P.eR-style

- - - -

Unrelated, there are at least thirty-five Venture Bro.'s (so far.) I don't want to show my work here because it would be spoiler.

- - - -

Seriously though: this is how "droids"(tm)(R) should talk, no?

- - - -

Also unrelated, the only speech act of H.E.L.P.eR's that I can understand in the entire series is when he and Brock are discussing poetry and he asks Brock, "Maya Angelou?". That's it. :-)

This reminds me of voice inversion, a scrambling method used by some analogue portable radios.

https://en.m.wikipedia.org/wiki/Voice_inversion

The last time I listened to these was ten years ago. I understood all of them just now.

Whoa. Spooky. I wonder if this works with a text spoiler also.