(PP1306) Comparing Speech Recognition with Remote Microphone Technology Using KEMAR and Voice-to-Text Testing

During the COVID-19 pandemic, traditional data collection methods were interrupted. With no in-person human subjects testing available, creative solutions were needed to continue research. An alternative method using voice-to-text, KEMAR, and remote microphone technology will be presented. The transcription application “Otter” was used due to its accuracy to provide responses. A Phonak M90 RT hearing aid was programmed for a flat 60 dBHL hearing loss. This was fit on an acoustic manakin, KEMAR, to represent a patient with a moderate hearing loss. Results for all remote microphones showed increased accuracy as SNR was increased supporting use of this protocol.

Summary: 

Purpose: In the past two years, researchers have experienced firsthand how a pandemic can halt in-person research. Over this time, research with in-person human subjects has been at times prohibited. This poster explains a creative solution allowing research to be conducted simulating human responses without participants. To test the accuracy of the setup, an experiment was conducted examining the benefit of remote microphones on hearing aid users. Previous research has shown a significant improvement when using a remote microphone in noisy environments (Tittle 2020, Thibodeau 2014). This warrants further research into new developments on remote microphone technology.

Methods: KEMAR represented a participant with hearing loss wearing a Phonak M90 RT hearing aid programmed to a 60 dBHL flat hearing loss on the right ear. Three different SNR conditions were performed using the remote microphones Roger Pen, Roger Select, SHARP 2+. A remote microphone was connected to the hearing aid and a speech in noise simulation using HINT sentences was played at the following SNRs: Quiet, +10, +15 dBSNR. A remote microphone was placed 1 foot and 0° azimuth from a speaker in a sound booth. KEMAR was placed 3 feet and 0° azimuth from the same speaker. The Hearing In Noise Test (HINT) sentences was used as stimuli. Babble noise was routed into a GSI 61 audiometer used for background noise which was delivered 90° azimuth to KEMAR. The transcription application “Otter” was open on the laptop and used due to its accuracy to provide responses. The output of KEMAR was connected to a Dell Latitude laptop that transcribed what was heard from the remote microphone into the Otter application. The speech was recorded on Otter and each word transcribed correctly was counted as one point. Word percent correct scores were given for the HINT sentence list. For each condition, the HINT sentence list 4 was presented three times.

Results: Preliminary data obtained showed similar performance across remote microphones that increased as SNR increased. At a 10dBSNR SHARP 2+, Roger Select, and Roger Pen performed at 52%, 47%, 52% respectively. When SNR was increased performance increased to 52%, 59%, 56% respectively. This validates the testing method.

Regardless of the pandemic, there is still research to be conducted and this setup offers an option for real world testing without participants present. The benefit of using this setup with KEMAR and voice-to-text is human subjects are not required. The setup can be modified and adapted for additional research studies. Additionally, this can be an option when there are difficulty recruiting participants or when testing is not available.

APA references:

Tittle, S., Thibodeau, L. M., Panahi, I., Tokgoz, S., Shankar, N., Bhat, G. S., & Patel, K. (2020). Behavioral Validation of the Smartphone for Remote Microphone Technology. Seminars in hearing, 41(4), 291–301. https://doi.org/10.1055/s-0040-1718714

Thibodeau L. (2014). Comparison of speech recognition with adaptive digital and FM remote microphone hearing assistance technology by listeners who use hearing aids. American journal of audiology, 23(2), 201–210. https://doi.org/10.1044/2014_AJA-13-0065