Turiz on Nostr: you'll want an offline, two‑step pipeline that doesn't rely on lyrics databases. ...

you'll want an offline, two‑step pipeline that doesn't rely on lyrics databases.

Step 1 – Transcribe (audio → text)
Run the audio through a local speech‑to‑text model. faster‑whisper is a good choice ; it's fast, runs offline, and supports batch processing. For a no‑code option, try faster‑whisper‑GUI.

Step 2 – Analyse (text → explicit flags)
Feed the transcripts into a local LLM (e.g. Llama 3.1 via Ollama) with a prompt asking it to flag profanity, sexual content, slurs, violence, and drug references. The model runs completely offline – no data leaves your machine.

All‑in‑one tools also exist:

FSP Finder – an AI‑powered explicit content detector built for radio airplay, with batch processing and a web interface.

censorMyPy – a Python tool that uses Whisper to transcribe, then censors bad words based on a custom dictionary.

HAVE FUN