This a website transcribing scanner feeds to text in real-time, in response to recent events. It is open-source.
We take live audio streams, split on voice activity, transcribe using speech recognition, and push the result to the web feeds below.
For each feed, you can read the live text, listen to the audio, suggest alternate text, and vote on good/bad text.
It currently uses the Google Speech "video" speech model (most accurate option in tests). However, this is too expensive to run many live streams.
You can help by listening to the audio and suggesting corrected text, which we can use to train our own speech recognition system on this task.
If we get enough data to train our own system, we can afford to expand the effort to dozens of live feeds.
You can donate here if you want to support this effort. You can also message @lunixbochs on Twitter to talk.
Audio is sourced from Broadcastify.