tools-core

Small, local-first utilities for media production, audio inspection, voice training research, and Gateway Forge authoring.

Open the project page → — project index, usage examples, contact details, and small interactive samples.

This repository contains source code only. It deliberately excludes recordings, lyrics, transcripts, model weights, checkpoints, generated media, caches, and third-party source trees.

Tool index

Area	What it does	Runtime	Platform	Writes files
audio-analysis/	Measures stereo beat layers, joins, onsets, pace, and prepares voice references	Python 3; NumPy/SciPy/soundfile for beat analysis; ffmpeg for corpus cutting	macOS/Linux	Some tools
audio-tag/	Tags untagged WAV, MP3 and FLAC files (Suno exports, bounces) with title, artist, album, year and track number from the folder layout, file names and creation dates	Python 3.8+, standard library	macOS verified; Windows/Linux unverified	Yes, in place (preview by default)
bundle-list/	Lists the files belonging to one asset: sidecars, photo pairs, multipart archives and frame sequences	Python 3.8+, standard library	macOS verified; Windows/Linux unverified	No
caption-check/	Finds subtitle timing and readability problems, with exact cues to inspect	Python 3.8+, standard library; optional fontconfig	macOS verified; Windows/Linux unverified	No
exact-duplicates/	Finds byte-identical copies across folders, accounting for overlapping roots and hard links	Python 3.8+, standard library	macOS verified; Windows/Linux unverified	No
file-identify/	Says what a file is from its bytes: the real format, a document's own heading, when it happened, where it came from. auto-sort's identifier, kept identical to it	Python 3 standard library; uses ffprobe, exiftool and OCR when installed	macOS/Linux/Windows	No
gateway-authoring/	Indexes measured signals, verifies a speech corpus, and seeds provisional Gateway Forge briefings	Python 3	macOS/Linux	Yes
piper-workflows/	Piper training, export, render, audition, and direct-inference wrappers for known upstream failure modes	Python 3 plus a compatible Piper training environment	Primarily macOS/Linux	Yes
remux/	Changes a media container by copying every supported stream; refuses encoding or track removal	Python 3.8+, FFmpeg and ffprobe	macOS verified; Windows/Linux unverified	A new copy
tray-icon/	Puts an icon with a menu in the menu bar, notification area or panel, each entry running a command. auto-sort's tray, kept identical to it	Python 3 standard library	macOS/Linux (seen); Windows (built to the documentation, not yet seen)	No
tts-research/	Produces Qwen3-TTS tensor, tokenizer, and speaking-pace ground truth	Python 3, MLX, mlx-audio, Transformers	Apple silicon macOS	Yes
site-capture/	Photographs the workshop's project pages whole, in both themes, repeatably	Python 3 and Chrome or Chromium; Pillow to verify and trim	macOS/Linux	Yes
site-fonts/	Moves a page's webfonts out of Google and into its own repository, keeping only the cuts it uses	Python 3 and Chrome or Chromium	macOS/Linux	Yes

The karaoke and narration renderer that used to live here as lyric-video-maker/ is now Protoke, its own cross-platform project.

Each directory has its own README with commands, dependencies, inputs, outputs, and caveats. Run commands from the repository root unless a README says otherwise.

Release check

python3 scripts/check_release.py

The check compiles every Python file, validates shell syntax, runs the standard library tests, and rejects private absolute paths, likely secrets, generated media, model artifacts, caches, and unexpectedly large files.

Extracted engines retain their upstream code and source-hash manifests. The apps are the source and these are copies: to bring every engine up to date with the app it comes from, checked and pushed in one step, run scripts/update-from-apps.sh with the apps checked out beside tools-core. The release tests check these with scripts/from-auto-sort.py --check and scripts/from-projects.py --check. Fix engine code in its parent project, then use the relevant copy script; see each tool's README. Remux integration tests exercise real multi-track media when FFmpeg and ffprobe are installed.

Dependency setup

The small utilities use Python's standard library. Optional dependency groups are documented in requirements/. Large ML stacks are not installed automatically because compatible versions depend on the model and hardware being investigated.

The wider workshop

• Gateway Forge — guided-session authoring and an experience journal.
• Voice Forge — measured Piper speech, pronunciation, and export.
• Protoke — lyric and narration video with a word-timed vector face.
• Media Preflight — check a finished file against a delivery target, and correct it without touching the original.
• siphon — fetch from a link, or convert what you already have, without re-encoding what it does not have to.
• auto-sort — empty the Downloads folder into structure it learns from what the files are, and undo any of it.
• tools-core — this repository.

Privacy and scope

• These tools take explicit paths and act only on what you name.
• No telemetry is added by these tools. The Piper wrappers explicitly disable ONNX Runtime telemetry before loading Piper.
• Copyrighted source audio and lyric sheets are inputs, not redistributable fixtures. Keep them outside this repository.

License

The repository is MIT licensed except for piper-workflows/, which is GPL-3.0-only because one export workaround adapts a small part of Piper's GPL implementation. Its license and attribution are included in that directory. Downloaded model repositories retain their own licenses and are not vendored here. See LICENSE for the exact boundary.