Anki Audio Quick Editor

Anki desktop add-on for quickly editing audio references from the note editor. It is optimized for short sentence-mining clips: trim edges, adjust speed, shorten long pauses, and automatically apply each edit as a new MP3 while leaving original media untouched.

Preview

<p> <img src="docs/assets/editor-prosody-playback.gif" alt="Inline prosody graph playback inside the Anki editor" width="49%"> <img src="docs/assets/editor-processing-chain.gif" alt="Applying audio processing actions from the Anki editor toolbar" width="49%"> </p> <p> <img src="docs/assets/browser-card-selection.gif" alt="Selecting cards in the Anki browser before batch audio work" width="49%"> <img src="docs/assets/editor-pitch-cursor.gif" alt="Inspecting the pitch cursor in the inline audio graph" width="49%"> </p> <p> <img src="docs/assets/editor-dark-prosody-playback.gif" alt="Inline prosody graph playback in Anki dark mode" width="49%"> </p>

What It Includes

• Inline Anki editor controls for fields containing [sound:...] references
• Inline prosody visualization with pitch, intensity, horizontal zoom, and a draggable playback start cursor
• ffmpeg-backed MP3 rendering for each inline edit action
• Silencedetect/Silero pause removal with optional denoise preprocessing and retained debug artifacts
• Non-destructive save flow that writes a new media file and updates the field reference
• Settings dialog and inline editor controls backed by committed Svelte/TypeScript bundles
• Config defaults, JSON Schema validation, and deep-merge migration support
• Unit tests, architecture tests, and real-runtime e2e tests
• scripts/dev.py for setup, checks, builds, release, and e2e execution
• .ankiaddon packaging via scripts/release.py

Requirements

• Anki 25.09 or later
• Python 3.13 as bundled by Anki
• Public release archives are thin. On first load, the add-on downloads the verified runtime pack for macOS arm64, macOS x86_64, or Windows x86_64.
• Optional advanced override: explicit ffmpeg_path still takes precedence over managed runtime discovery
• Locked bundled praat-parselmouth and NumPy wheels for pitch/intensity analysis on supported platforms; the graph analyzer still falls back to ffmpeg-decoded PCM if Praat is unavailable
• Node.js 18+ for editing or rebuilding the settings/editor frontend bundles

Quick Start

python3 scripts/dev.py setup
python3 scripts/dev.py runtime-install
python3 scripts/dev.py check
python3 scripts/dev.py test-e2e

The local development add-on ID is 1000000002.

Development

• Runtime package: addon/anki_audio_quick_editor/
• Settings and editor frontend source: settings_ui/
• Build the committed frontend bundles: python3 scripts/dev.py build
• Open the settings dialog from Anki: Tools -> Anki Audio Quick Editor -> Settings
• Edit audio from Anki by focusing a field containing a supported sound reference such as [sound:filename.m4a]; edits are saved as new MP3 files.

Release

Runtime packs and thin add-on archives are released separately. Build and publish a runtime release only when native tools or model files change. The canonical release checklist is in DEVELOPMENT.md.

python3 scripts/dev.py release-assets verify --target all
python3 scripts/dev.py release-runtime build --runtime-version 1.0 --target all
python3 scripts/dev.py release-runtime upload --metadata runtime_release.lock.json
python3 scripts/dev.py release-runtime verify --metadata runtime_release.lock.json

Normal public add-on releases are thin and consume the tracked runtime_release.lock.json metadata:

python3 scripts/dev.py vendor-wheels verify
python3 scripts/release.py --target all --verify-runtime-urls
python3 scripts/dev.py release-smoke dist/anki-audio-quick-editor-<version>.ankiaddon

Inspect the archive manifest before publishing: it must point at runtime-vN, not the add-on tag. The release script also verifies packaged wheels against addon/anki_audio_quick_editor/vendor/wheels.lock.json. GitHub add-on releases should upload only the .ankiaddon; AnkiWeb should receive the same smoke-tested archive.

Before publishing a public release, run native acceptance on macOS arm64, macOS x86_64, and Windows x86_64. That includes checking that Anki's bundled CPython 3.13 imports the packaged praat-parselmouth and NumPy wheels from the add-on-local user_files/python_vendor/<platform>/ cache, then verifying Graph and Pitch Hum run without the ffmpeg/PCM fallback warning.

Similar projects

• Onsei by Didier Marin works nicely on the web, he also has an anki plugin for comparing pitch
• kotu.io by kezi with backup of its early version by kuuube - an amazing set of learning tools, inlcuding the famous minimal pairs trainer
• migaku's pitch trainer
• pitchaccentapp by checkempty - another web tool for pitch comparison
• Praat - an ancient tool for researchers. We all use algorithms they developed
• YuTone - an app for chinese, somehow super fast, processes pitch in real time
• Aomi Japanese records and compares your Japanese pitch accent
• Tone perfect and companion reading-feedback app - chinese only,
• Ka Chinese Tones - kotu-like tool for chinese tones
• Aaron's Vietnamese toolset
• OJAD some tools for Japanese pitch, including pitch of sentences
• JPitch - word pronunciation analysis for japanese, feedback on mora-by-mora level!
• MandaTone - Chinese tones app
• CanTone - comparing your cantonese tones against model
• Vietnamese Tones ios app for recording your tones
• Pho speak big vietnamese course that includes pronunciation feedback
• Cracking languages Tons of tools for tonal languages from a tonal polyglot
• Syllable
• Thai language tones