Transcribe audio from apps, files and your device's microphone

Hi HN,

I built a free Android transcriber that has no ads or extra downloads to start transcribing without any limit. It’s a simple app, all things considered; however, what I would like the community’s input on is how I can improve the accuracy.

I currently use the C++ TensorFlow Lite library to load the base(yup base, not tiny) English Whisper model into the app. This is passable for the most part, with about 88%–93% accuracy based on my research and experience using the app. But these newer models, like the ones used in the basic Google Recorder, hover around 95%–96% accuracy (to simplify things).

I know that as a solo developer, I won't have access to those proprietary Google models. However, if anyone is familiar with an easily accessible model with higher accuracy than the base Whisper model that is roughly the same size (or smaller), I would love to hear about it for future releases. The model doesn’t necessarily need to be in a TensorFlow Lite format, I’m willing to redo the implementation if the accuracy and model size offer a significant improvement.

Of course, any additional feedback regarding other aspects of the app would be greatly appreciated as well.