Add CLI progress reporting: Running detect/trim used to sit completely silent through the whole pipeline, including the ASR step -- the slowest part on CPU and the one most likely to actually run -- with no indication anything was happening. Found via a direct user complaint: 'it would be nice if the cli tool emitted some sort of progress vs sitting there blankly for unknown time.' detect.detect() now takes an optional progress callback, called with a short message at each major stage (fetching audio, official SRT, ASR, reference correlation, sibling-episode fetching); defaults to a no-op so existing callers and tests are unaffected. cli.py wires this to a single, in-place-overwriting stderr line (so it never pollutes piped/ redirected stdout) for detect, sweep, and random-sample. Also documents the pre-2015 outro-wording fix (markers.json, already committed separately) in docs/DESIGN.md's Notable fixes, with a screenshot showing the actual speech-over-music overlap in Audacity that motivated it.
hpr-intro-stripper
Extracts the "inner" audio from a Hacker Public Radio episode — the host's originally submitted recording, with HPR's own added intro and outro removed — for one episode or in batch.
It never ships or assumes a local copy of the HPR archive: it fetches whatever metadata, audio, and transcripts it needs from the live HPR site itself, and never touches the original files.
For the full technical background (why this isn't trivial, how HPR's own pipeline works, the detection strategy, benchmarks, and known limitations), see docs/DESIGN.md.
Motivation
From Ken Fallon's comment on hpr4761 :: HPR Community News for October 2026 (transcript, lightly corrected for clarity):
That had me thinking, actually Lee — a lot of the original episodes on HPR have the HPR intro and outro on them, and I was just wondering: would there be a way where we could get the original source material and strip out the old intro and outro, so that we just have the audio for that, and then update them as time goes on?
Requirements
- Python 3.9+
ffmpeg/ffprobesqlite3andawk(for fetching/converting HPR's metadata dump)numpy(pip install numpy)- Optional, only used as a fallback when HPR hasn't published an official
transcript for an episode:
openai-whisper(pip install openai-whisper), run with thetinymodel on short clips only — never a whole episode, never requires a GPU.
No step requires a GPU or a specific speech-recognition setup.
How it works, briefly
HPR's intro is a per-episode, auto-generated spoken announcement (title, host, date, summary) over a music bed; the outro is a single fixed clip. Detection tries, in order: HPR's own official transcript/subtitle file (if published for that episode), a metadata-duration cross-check, local speech recognition on a short clip, then audio correlation against known reference clips — whichever signal actually works for a given episode, recorded alongside the result. Episodes with no reliable signal, or an unusual/No-Derivatives license, are flagged for manual review rather than guessed at. See docs/DESIGN.md for the full reasoning.
Usage
pip install numpy openai-whisper # openai-whisper only needed for the ASR fallback
# Detect cut points for one episode (fetches audio/metadata as needed, caches in ./cache)
python -m hpr_stripper.cli detect --episode 3700
# Detect a range
python -m hpr_stripper.cli detect --range 3700-3720
# Prefer a different audio format (default: mp3, then ogg, then opus)
python -m hpr_stripper.cli --format ogg,opus detect --episode 3700
# Produce trimmed output for everything detected so far with a usable result
python -m hpr_stripper.cli trim --out-dir ./trimmed
# Random, era-stratified spot check
python -m hpr_stripper.cli random-sample --n 12
# Full-archive detection pass -- safe to interrupt (Ctrl+C) and resume by re-running
python -m hpr_stripper.cli sweep
# Summary of what's been processed so far
python -m hpr_stripper.cli report
All intermediate state (downloaded audio, the metadata database, transcripts,
detection results, trimmed output) lives under ./cache/ and ./trimmed/
by default, both gitignored. Nothing here is uploaded or shared anywhere by
this tool.
License handling
Every episode's own Creative Commons (or other) license is read from HPR's metadata and carried through to the output. Episodes under a No-Derivatives license are never trimmed for distribution by default (see docs/DESIGN.md). Trimmed output is accompanied by a small JSON sidecar recording the host, episode number, date, and license, since the intro being removed is also HPR's mechanism for crediting hosts under CC-BY-SA.