PP1324 - Voice-to-Text Speech Recognition with Over-the-Counter and Traditional Hearing Aids

With FDA approval of over-the-counter (OTC) devices, basic amplification has become more accessible for those with mild hearing loss. To compare OTC devices with traditional amplification with multiple noise reduction features, a protocol was developed using a voice-to-text application, Otter.ai, and KEMAR. When output was matched among three devices (OTC-Lucid, Phonak P30, and P90 aids) that were programmed for a flat 40 dB HL loss, speech recognition in steady speech-shaped noise increased as signal-to-noise level increased from -5 to +10 dB and was best for the highest level technology as expected.

Summary:

A GRAS KEMAR head and torso simulator represented a person with hearing loss. KEMAR was fitted with 3 hearing devices programmed to a flat 40 dBHL sensorineural hearing loss on the right ear to simulate a mild hearing loss. One Phonak P30 RT receiver-in-the-canal (RIC), one Phonak P90 RT RIC, and a Lucid Engage over-the-counter device using a RIC were assessed. The Phonak devices were adjusted in their proprietary software with all noise reduction features maximized. The OTC was adjusted in the Lucid app running on an iPhone to closely match the output of the Phonak hearing aid (+ or – 5 dB). Electroacoustic analysis was performed on the devices using a Verifit 2 and Fonix 8000. KEMAR was placed 3 feet and 0° azimuth from a loudspeaker in an audiometric soundbooth. The speech-shaped noise from a GSI 61 audiometer was presented from a loudspeaker 8 feet and 180° behind KEMAR.

A modified sentence list from the Hearing In Noise Test (HINT) was presented at 65 dB SPL as measured at the hearing aid microphone on KEMAR’s ear. The modified sentence used a carrier phrase of the repeated word “boy” before the delivery of the target sentence. To assess the accuracy of each device’s processing, the output from the Zwislocki coupler in KEMAR’s ear was sent to a laptop and recorded on Otter.ai. Word percent correct scores were determined for the modified HINT sentences. Four conditions were assessed: +10, 5, 0, -5 dBSNR. The modified HINT sentence list 4 was presented three times for each condition. The final score for each condition was the average number of words correctly transcribed across the three trials. The standard deviation between the three trials was calculated. In conditions with deviations greater than or equal to 25 a 4th trial was run and the final score was calculated using the average of 4 trials.

Analysis was completed using a two-factor ANOVA without replication. There was statistical significance in speech-in-noise scores between hearing devices and between listening conditions. Differences in factors were not due to chance. A post-hoc analysis revealed a significance between the P30 and P90 devices. No other differences were noted between devices. A post-hoc analysis revealed a significance between all listening conditions except between -5 and 0 SNR. Electroacoustic findings between the OTC and traditional hearing aids yielded differences in several measures.

This preliminary data provides evidence that an OTC can be a viable option for those with mild hearing loss who also express concerns with hearing in noise. It is important to consider that traditional hearing devices offer specialized adjustments from hearing healthcare professionals that can optimize devices specifically for each person. OTCs provide gross adjustments that can be changed by an individual, but may not be maximized to their full potential without the guidance of a hearing healthcare professional. Further data collection on user adjustments compared to optimal adjustments on OTCs should be compared in the future. Additional limitations of using one device per manufacturer and steady-state noise were noted.