standards of speech intelligibilty for the hearing impaired?

[Arne Leijon <arne.leijon@xxxxxxxxx>]

There have been a long series of attempts to modify the standard SII  
procedure ANSI S3.5(1997)
to include some basic aspects of hearing loss in predictions of speech  
intelligibility
for practical purposes, for example

@article{Pavlovic1986,
	Author = {Pavlovic, C.V. and Studebaker, G.A. and Sherbecoe, R.L.},
	Journal = jasa,
	Pages = {50-57},
	Title = {An articulation index based procedure for predicting the  
speech recognition performance of hearing-impaired individuals},
	Volume = {80},
	Year = {1986}}

@article{Ching1998,
	Author = {Ching, Teresa Y C and Dillon, Harvey and Byrne, Denis},
	Journal = jasa,
	Number = {2},
	Pages = {1128-1140},
	Title = {Speech recognition of hearing-impaired listeners:  
predictions from audibility and the limited role of high-frequency  
amplification},
	Volume = {103},
	Year = {1998}}

@article{Ching2001,
	Author = {Ching, Teresa Y C and Dillon, Harvey and Katsch, Richard  
and Byrne, Denis},
	Journal = {Ear and Hearing},
	Number = {3},
	Pages = {212-224},
	Title = {Maximising effective audibility in hearing aid fitting},
	Volume = {22},
	Year = {2001}}

The SII approach itself, and those modifications, are of course all  
too simple to give a realistic
model of all the various physiological factors that can reduce speech  
intelligibility.
Nevertheless, the amazing thing is that even those very simple models  
actually account
astonishingly well for some results of speech tests in noise, e.g.

@article{Magnusson1996a,
	Author = {Magnusson, L.},
	Journal = {Scandinavian Audiology},
	Number = {4},
	Pages = {215-222},
	Title = {Predicting the speech recognition performance of elderly  
individuals with sensorineural hearing impairment},
	Volume = {25},
	Year = {1996}}

For some special purposes, it may even be more reliable
to use the acoustical prediction than to use a real speech-recognition  
test!

@article{Magnusson2001,
	Author = {Magnusson, Lennart and Karlsson, Mia and Leijon, Arne},
	Journal = {Ear and Hearing},
	Number = {1},
	Pages = {46-57},
	Title = {Predicted and measured speech recognition performance in  
noise with linear amplification},
	Volume = {22},
	Year = {2001}}

Still we are probably far away from any physiologically realistic  
standard.

However, given any (more or less realistic) auditory model, it is  
always possible to estimate
the amount of speech information that can be transmitted through the  
model, e.g.

@inproceedings{Stadler2007,
	Address = {Antwerpen, BE},
	Author = {Svante Stadler and Arne Leijon and Bj{\"o}rn Hagerman},
	Booktitle = {Interspeech 07},
	Title = {An Information theoretic approach to estimate speech  
intelligibility for normal and impaired hearing},
	Year = {2007}}

This approach is equivalent to testing how well an ideal automatic  
speech-recognition system can perform, given the
speech signal output from the auditory model.

Best wishes,
Arne Leijon