NOTE: Available at www.fspfinder.com (HF link is deactivated)

FSP (Foul Speech Pattern) Finder is a useful tool for preparing music files for radio airplay by detecting and automatically censoring explicit content. We use a custom version of OpenAI's automatic speech recognition model Whisper, which we fine tuned on over 100 hours of music vocals, to transcribe uploaded music files (with timestamps for each word). We then search for common explicit words (e.g., curse words, racial slurs, etc.) in the transcript. The vocals stem of the track is separated using demucs, then muted at the identified times, to produce a high quality radio-friendly version of the uploaded track(s).

Our tool comes with an easy to use web interface built in Gradio. The tool can process files one at a time or in batches, and the web interface allows the user to view the full transcript of each track along with the words that will be censored, before downloading the edited files.