All marketing departments seem to want to jump on the ChatGPT AI bandwagon. Well, IMO, saying “we use AI to improve your hearing in noise” is rather vague. Reminds me of when JC Whitney promised “up to 25% improvement in gas mileage” for a whirly-go-round device under the carburetor. 0% is up to 25%.
We have been testing some new video AI software still under development. We are using it in conjunction with surveillance camera software (Network Video Recorder). The camera sends video streams to the NVR and the NVR (based on detection criteria) sends images to the AI software for evaluation. There are various “models” of images containing objects to be used for identification. These files get quite large. Some have animals, some have license plates, some have people and vehicles, etc. We use a model that contains “persons” and “vehicles”. We can define vehicles if we want to, such as cars, trucks, forklifts, etc. We also tell it criteria such as % confidence, how many samples to evaluate and how often (like 100 milliseconds) between samples. It is very complex. Using the video processor in the NVR, recognition times are in the 60-80 msec range if it recognizes the image. It makes mistakes, sometimes it thinks our metal stairway is a truck, depending on shadows. Very few false alerts. Without the AI package, vehicle headlights cause many false triggers, with AI almost none. One user claims that his cat is mad because the AI says it is a dog . Fine for surveillance, but I don’t want the same performance in health care.
We all want better hearing in noise. AI sounds like a perfect solution until you consider that a lot of that noise is from voices that we don’t want to hear versus voices we do want to hear. I suppose you could provide a model of the voices that you do want to hear, for AI to examine. But, what if it introduces delays in the half second range?
My Omnia’s are doing quite well with forward focus, (up to 100%).