[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Rationale for Speech Tests

Hi Enrique,

I can only respond to Q2:
In "Krenmayr A, Qi B, Liu B, Liu H, Chen X, Han D, Schatzer R, Zierhofer CM. Development of a Mandarin tone identification test: sensitivity index d' as a performance measure for individual tones. Int J Audiol. 2011 Mar;50(3):155-63" we have used signal detection theory to model psychometric functions of Mandarin tone identification. In particular to model the performance of individual tones, for which the percent correct scores have turned out to be considerably biased.

Mandarin comprises four tones, when a listener responds "tone 1" in all presentations, the corresponding percent correct for tone 1 would be 100% and for all other tones it would be 0. Every listener seemed to have a bias towards one favorite tone (even though not as extreme as in the example above), so the psychometric functions based on percent correct had all sorts of lower asymptotes. Nevertheless, only the value of 25%, the guess rate, could be the true lower asymptote.
Calculating a d' measure, we were able to get rid of the response bias and found nicely linear psychometric functions for d' over SNR.

Kind regards,

Andreas Krenmayr, Dr. Dr.
Team Leader Studies/Research&Development                                          
MED-EL Medical Electronics
Fuerstenweg 77a
6020 Innsbruck, Austria
Tel +43512288889860 
Mobile +4369918888860 
Fax +43512288889490


-----Original Message-----
From: AUDITORY - Research in Auditory Perception [mailto:AUDITORY@xxxxxxxxxxxxxxx] On Behalf Of Enrique A. Lopez-Poveda
Sent: Friday, April 20, 2012 10:39 AM
To: AUDITORY@xxxxxxxxxxxxxxx
Subject: Rationale for Speech Tests

Dear all,

I would appreciate your help on the following two questions.

Q1. Conventional speech perception tests are run using a fixed speech (or noise) level and varying the speech-to-noise ratio (SNR); the implicit assumption is that the speech reception threshold (SRT) is independent of the absolute speech or noise levels. I would appreciate some references that proof this statement and justify this procedure (PDFs of the appropriate references would be even better).

Q2. Can speech psychometric functions (that is, functions representing percent correct identification versus SNR) be modeled using signal detection theory? I would also appreciate references.

Thank you so much in advance,

- Enrique

Enrique A. Lopez-Poveda, Ph.D.
Instituto de Neurociencias de Castilla y Leon Universidad de Salamanca Calle Pintor Fernando Gallego 1, 37007 Salamanca, Spain
Phone: (+34)923294500 ext. 1957; Fax: (+34) 923294750 http://web.usal.es/ealopezpoveda
Skype: ealopezpoveda