OpenAI Whisper Keyboard

Kai's Soapbox

About this app

The OpenAI Whisper Voice Keyboard by Kaizo Co is a powerful speech recognition keyboard that unlocks the power of OpenAI's Whisper Speech Recognition. With Whisper, you can unlock the power of multilingual speech recognition, speech translation and language identification But right now we are only using the tiny English model, which is small and lightweight. So, you can enjoy faster and more accurate voice input. With Whisper, you can quickly and accurately input text with just your voice. Try the OpenAI Whisper Voice Keyboard by Kaizo and Co today! We make no guarantee as to the level of your resulting satisfaction.

Ratings and reviews

It's not perfect, but there's nothing out there that is. It's now my main keyboard app for the past six months or so. My favorite thing about it is that I can pause for as long as I want and then just continue speaking and It works just fine, unlike the Google keyboard (Gboard). The most annoying thing is that I have to switch to another keyboard if it makes a mistake I need to go edit. Edit: Found that FUTO uses the version 3 of the whisper model and so I use that now.

It's very basic and does just what it says, but it gets a lot of words wrong because it's using a tiny model, Even with Max threads. I have a lot of space on my phone, so I would rather download and use the V3 model if it was an option. A more advanced keyboard would follow gboards example and type during silence but use the whole text for context in following chunks. Even better would be to let me select words and choose other highly scored alternatives.

It's okay, but has some issues. I'm finding that too many words drop out completely, and some are just mistaken. I think it may be the language model that's being used in the app... United States English. Unfortunately, it doesn't allow the user to select any different language or other form of English. I speak with a New Zealand English accent. I look forward to an improved version with the option to choose the language and model size. Currently, it's dropping too many words to be useful.