[AUDITORY] Automatic speech recognition in Presentation (Neurobehavioral Systems)

[Moritz Wächtler <moritz.waechtler@xxxxxxxxxxxxxxxxxxxx>]

Dear list,

does someone of you use the built-in automatic speech recognition 
feature of Presentation (Neurobehavioral Systems) in their research?

We currently try to set up a speech intelligibility test in which the 
subjects are presented with a sentence and are then asked to repeat what 
they just heard while speaking into a microphone. Our goal is to 
determine the verbal response delays, that is, the time between the 
offset of the auditory stimulus (the sentence) and the onset of the 
subject's verbal response. The onset of the subject's response can in 
many cases be detected by simply thresholding the sound level at the 
microphone. However, this method fails when the subject produces some 
other sounds such as breathing or non-speech utterances ("Um", "Hm", 
...).

So, in an attempt to make the response onset detection more robust, we 
want to use Presentation's automatic speech recognizer with a dictionary 
only containing the test's word material.

However, so far, the results of this approach are not satisfactory. 
Among other things, the speech recognizer often misses words, causing 
biased response delay estimates.

Does anyone of you have experiences with this or similar methods?

Thanks in advance for any advice.
Best regards,
Moritz