(This is all about the mobile application)
Whisper is trained on data that includes details about captions, ex: [Music] or “Thank you for watching!” (Even though it is captions, I included “Thank you for watching!” because it was mostly trained on silent audio, like in a YouTube video with only captions and no audio) which greatly effects the output. This is prevalent in all of whisper, but effects the mobile application more, not sure on exact numbers though, but it is most likely because, If I am correct on how it is processed, it uses the Whisper API and not directly OpenAI API. This is going off of what I have read in the forums, example: /t/why-whisper-accuracy-is-lower-when-using-whisper-api-than-using-openai-api
Haptic feedback is based on the devices percentage, not it’s “Low Power Mode” state. If low power mode is not on but the device is below 20%, it will not be able to be turned on.
I already listed the issues on whisper above, but adding on to that, after using the voice feature, whisper & whatever they use for TTS, will randomly add text even after you are done talking, which is part of the silence being captioned.