Article
Determination of temporal thresholds for the detection of speech from spectrally-matched noise
Search Medline for
Authors
Published: | September 3, 2020 |
---|
Outline
Text
In acoustic scene analysis, humans separate and recognize ambient sounds that arrive at their ears to distinguish between important and negligible signals. For the identification, the spectral and temporal characteristics of the signals are analysed. Since the acoustic scene is constantly changing, it is important that the detection of different signals takes place as quickly as possible. Thus, for example, a suddenly approaching car in road traffic can be detected in time.
Within this contribution, the temporal thresholds for the detection of different speech samples were examined. This means the minimum length of a speech signal required for a correct detection was measured. To this end, a 2-AFC procedure was used where a spectrally-matched noise signal was presented as an alternative. A study was conducted with 30 elderly subjects whose hearing was within the WHO classes 0 to 3. Ten speech samples with different initial phonemes were considered and individual noise signals were created for each sample and duration.
The results show a significant influence of the initial phoneme on the temporal detection thresholds. However, no general relation between hearing loss and detection thresholds was found. Since hearing impaired people are less able to process the temporal fine structure, the results suggest that the envelope of a speech signal is used for the speech detection rather than the temporal fine structure. In addition, a correlation between age and cognition with the detection thresholds was found, suggesting an aging process of auditory processing.