Evaluation of a "direct-comparison" approach to automatic switching in omnidirectional/directional hearing aids.

BACKGROUND: Hearing aids today often provide both directional (DIR) and omnidirectional (OMNI) processing options with the currently active mode selected automatically by the device. The most common approach to automatic switching involves "acoustic scene analysis" where estimates of various acoustic properties of the listening environment (e.g., signal-to-noise ratio [SNR], overall sound level) are used as a basis for switching decisions.

PURPOSE: The current study was carried out to evaluate an alternative, "direct-comparison" approach to automatic switching that does not involve assumptions about how the listening environment may relate to microphone preferences. Predictions of microphone preference were based on whether DIR- or OMNI-processing of a given listening environment produced a closer match to a reference template representing the spectral and temporal modulations present in clean speech.

RESEARCH DESIGN: A descriptive and correlational study. Predictions of OMNI/DIR preferences were determined based on degree of similarity between spectral and temporal modulations contained in a reference, clean-speech template, and in OMNI- and DIR-processed recordings of various listening environments. These predictions were compared with actual preference judgments (both real-world judgments and laboratory responses to the recordings).

DATA COLLECTION AND ANALYSIS: Predictions of microphone preference were based on whether DIR- or OMNI-processing of a given listening environment produced a closer match to a reference template representing clean speech. The template is the output of an auditory processing model that characterizes the spectral and temporal modulations associated with a given input signal (clean speech in this case). A modified version of the spectro-temporal modulation index (mSTMI) was used to compare the template to both DIR- and OMNI-processed versions of a given listening environment, as processed through the same auditory model. These analyses were carried out on recordings (originally collected by Walden et al, 2007) of OMNI- and DIR-processed speech produced in a range of everyday listening situations. Walden et al reported OMNI/DIR preference judgments made by raters at the same time the field recordings were made and judgments based on laboratory presentations of these recordings to hearing-impaired and normal-hearing listeners. Preference predictions based on the mSTMI analyses were compared with both sets of preference judgments.

RESULTS: The mSTMI analyses showed better than 92% accuracy in predicting the field preferences and 82-85% accuracy in predicting the laboratory preference judgments. OMNI processing tended to be favored over DIR processing in cases where the analysis indicated fairly similar mSTMI scores across the two processing modes. This is consistent with the common clinical assignment of OMNI mode as the default setting, most likely to be preferred in cases where neither mode produces a substantial improvement in SNR. Listeners experienced with switchable OMNI/DIR hearing aids were more likely than other listeners to favor the DIR mode in instances where mSTMI scores only slightly favored DIR processing.

CONCLUSIONS: A direct-comparison approach to OMNI/DIR mode selection was generally successful in predicting user preferences in a range of listening environments. Future modifications to the approach to further improve predictive accuracy are discussed.