How auto-detect and remove speech/dialogue in music radio recordings?

Question

My goal is to review and selectively remove speech, especially known adverts/commercials, from numerous recordings of a community music radio station (that I work for).

Are there any DAW plugins (eg VST) or apps or projects that could assist with this? Preferably (but not necessarily) macOS or macOS-buildable.

For example, a DAW plug-in to generate an "indicator track" alongside a subject track, or to insert marker spikes or tones (eg to a copy of that track).

Alternatively, are there any stand-alone options?

(I initially asked at superuser, where I was advised to instead post here).

Answer 1

I don't know if this would be convenient enough, but one option could be using whisper.cpp with some small (= relatively fast) English-only model (I'm taking assumption that this station is using English, from my tests multi-language models have worse performance), transcribing recordings with maybe JSON or CSV with timestamps as output. It can be filtered then with known keywords, some simple filtering GUI might be needed. I'm guessing you would need to delete audio starting from the of files to keep timestamps in sync.