[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: phoneme recognition and stimulus length
Tóth László wrote:
> On Mon, 19 Mar 2001, Jont Allen wrote:
> > I dont know if this is what you are getting at, but have you looked at the
> > paper by Sadaoki Furui "On the role of spectral transitions for speech perception."
> > JASA, Oct. 1986, page 1016+
> It partly answers my questions, as they used truncated syllables as
> stimuli. But I would be more interested in the other direction:
> is there any additional gain in recognition performace if we use stimuli
> LONGER than a syllable?
As I mentioned last month, the work of Cyma Van Petten (JASA, page 2643,
Nove 2000, #5, pt 2, vol 108) is relevant here. She showed that there is a point
in time where you resolve the word/sound, and that adding more time or sound
does very little to the score. At least that is how I remember what she said.
If you search for her name with google you will find her home page with
references. This is a large body of work. The title of her paper at ASA is
"Time course of word id adn semantic integration in spoken language."
> What made me wonder about this is the "backwards recognition masking"
> experiments of Massaro (unfortunately, I don't have the original papers,
> only a half-page review in a Ph.D. thesis by Brian Kingsbury). Their
> results say that masking has no effect if the target is longer than a
> syllable or if there is at least a syllable-long silent interval between
> the target and the masker. I would
> need a reinforcement of these results, but possibly from the opposite
> direction (i.e. not how recognition deteriorates from backwards masking
> but how recognition improves from "forward helping" - so to say).
> Laszlo Toth
> Hungarian Academy of Sciences *
> Research Group on Artificial Intelligence * "Failure only begins
> e-mail: email@example.com * when you stop trying"
> http://www.inf.u-szeged.hu/~tothl *
Jont B. Allen
AT&T Labs-Research, Shannon Laboratory, E161
180 Park Ave., Florham Park NJ, 07932-0971
973/360-8545voice, x7111fax, http://www.research.att.com/~jba