| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
| |
scripts/ratio-transcribe runs whisper plus speaker diarization on my own transcription host over an ssh job queue, with a local worker as the fallback when the host is unreachable. It needs no API key, so its descriptor carries nil for the auth host and the env var. The process environment passes through untouched. SPEAKERS=N in the environment pins the diarizer's speaker count.
The interactive switcher used to carry its own copy of the backend list, so a backend added to the descriptors was reachable only by setq. It now derives its choices from the descriptor alist, and a test pins the two to each other.
The script's own behavior (argument validation, the job-id hash, the remote and local branches) is tested in the repo it came from. A bats mirror of those cases is tracked separately.
|
| |
|
|
|
|
| |
make-process :stderr was a file path, but :stderr takes a buffer, so Emacs made a buffer named after the path instead of writing the file. The "Errored. Logs in <file>" notification pointed at a log with no error text, and that hidden buffer leaked one per transcription.
I now route stderr through an explicit, erased buffer: passed to :stderr, threaded to the sentinel, drained into the log, then killed. That keeps stderr off the stdout :buffer, so the transcript stays clean.
|
| |
|
|
| |
The video flow extracts audio to a temp .mp3 in /tmp, then derived the .txt/.log from that temp path, so transcripts landed in /tmp and died on reboot, contradicting the "alongside the source" docstring. Thread the source video through cj/--start-transcription-process as an optional output base so the outputs derive from it (talk.mp4 -> talk.txt beside the video). Audio is unchanged. New regression test pins the output base to the source video.
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
Phase 2 of the load-graph project. I fixed the seven hidden dependencies the classification surfaced, so each module declares what it uses instead of relying on init order.
- system-defaults now requires host-environment and user-constants at runtime. They were eval-when-compile only, but env-bsd-p and user-home-dir are read at load, so the compiled module couldn't load standalone.
- custom-buffer-file, dev-fkeys, calendar-sync, and video-audio-recording require keybindings and drop their (when (boundp 'cj/custom-keymap) ...) shims. The shim silently dropped the C-; binding when the module loaded before keybindings. The explicit require makes the dependency real.
- flycheck-config and mail-config require keybindings for their cj/custom-keymap bindings (a use-package :map and a direct keymap-set).
- Removed a dead eval-when-compile (defvar cj/custom-keymap) in transcription-config; nothing there used the variable.
No init.el load-order change. keybindings and the foundation modules already load before these, so the requires are no-ops at startup and only fix standalone and test loading.
I verified each fix with a fresh emacs --batch (require 'X), then swept all modules standalone: every one loads or fails only with a clear missing-package message. Full make test, make validate-modules, and an init smoke all pass. Module headers and the inventory's hidden-dependency section are updated to mark the seven resolved.
|
| |
|
|
|
|
| |
Final classification batch: the last 19 modules — linear-config, local-repository, lorem-optimum, mail-config, markdown-config, music-config, pdf-config, quick-video-capture, reconcile-open-repos, restclient-config, slack-config, system-commands, telega-config, tramp-config, transcription-config, video-audio-recording, vterm-config, weather-config, wrap-up. I annotated each header, added a Batch 9 table to the inventory, and extended the validation allowlist. 101 of 102 modules are now classified; only elfeed-config remains, deferred on its test fix.
Two more hidden dependencies turned up. video-audio-recording uses the boundp shim for its C-; r binding, and mail-config registers C-; e directly without requiring keybindings, so it errors standalone rather than degrading. Both recorded for Phase 2.
|
| |
|
|
|
|
|
|
| |
The auth-source-search + funcall-the-secret block was copied four times: calendar-sync--calendar-url, cj/auth-source-secret (ai-config), cj/--auth-source-password (transcription), and cj/slack--get-credential. Each searched authinfo, pulled :secret, and called it when the netrc backend returned a function.
I pulled that into cj/auth-source-secret-value in system-lib (a leaf, so calendar-sync doesn't have to depend on ai-config and drag in the gptel stack). It takes an optional user and returns the secret or nil. The four callers now delegate to it: ai-config layers its required-secret error on top, and the others keep their nil-on-miss behavior. With the direct auth-source-search calls gone, I dropped the now-unused (require 'auth-source) from transcription, slack, and calendar-sync. The helper's autoload covers it.
The transcription tests that exercise the delegated path stay green, and the primitive and the error wrapper get their own tests.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Pressing `T' in dired/dirvish on an audio file already transcribed
it; on a video file it bounced with "Not an audio file". Real
recordings ship as .mp4 / .mkv at least as often as raw .m4a, so
the one-key flow ended at the wrong place.
Pipeline now:
- audio path -> direct into `cj/--start-transcription-process'
(unchanged).
- video path -> async ffmpeg extracts the audio track to a temp
.mp3 under `temporary-file-directory' (libmp3lame, VBR q:a 4,
~165kbps -- right size for speech, accepted by every backend),
then transcribes that file with the temp marked for cleanup
after the transcription sentinel fires.
Surface changes:
- `cj/video-file-extensions' added to user-constants.el (mp4, mkv,
mov, webm, avi, m4v, wmv, flv, mpg, mpeg, 3gp, ogv).
- New predicates `cj/--video-file-p' / `cj/--media-file-p'.
- New `cj/--extract-audio-from-video' (async ffmpeg with success
callback; surfaces `cj/--notify' on failure; user-errors if
ffmpeg isn't on PATH).
- `cj/--start-transcription-process' gains optional `cleanup-file'.
Sentinel deletes it after the existing logic runs. Backwards
compatible -- the audio flow doesn't pass it.
- `cj/transcribe-audio' renamed to `cj/transcribe-media' (dispatcher
on audio vs video). `cj/transcribe-audio-at-point' renamed to
`cj/transcribe-media-at-point'. Both old names kept as
`defalias' so M-x history and any external references still work.
- `T' in dired-mode-map + dirvish-mode-map points at
`cj/transcribe-media-at-point'.
- Module commentary USAGE block updated.
15 new ERT tests in `tests/test-transcription-video.el' cover the
predicates (happy/boundary/error), ffmpeg invocation (correct args
+ missing-ffmpeg path), the dispatcher (audio direct, video via
extraction, non-media rejected), the aliases, and the T binding.
One existing test in `test-transcription-status-and-commands.el'
updated to stub the new delegate name.
Verified locally that ffmpeg is on PATH with libmp3lame, and that
the exact arg list my code uses produces a valid MP3 from a
synthetic test video.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The transcription menu wasn't earning its top-level keymap slot --
the commands (transcribe-audio, switch-backend, view-transcriptions,
kill-transcription) are run rarely enough that `M-x' is fine. Drop
the `cj/transcribe-map' keymap, its `(keymap-set cj/custom-keymap
"T" ...)' binding, and the which-key labels. Commands stay
callable by name.
That frees `C-; T' for telega, where the mnemonic actually fits.
Move the launcher from `C-; G' to `C-; T'. Update the
which-key label, the module commentary, and the keymap-binding
test assertion. The dashboard `g' single-letter binding stays put
-- `t' there is vterm, so dashboard letters and the global
`C-;' prefix don't share a key space anyway.
|
| |
|
|
|
|
|
|
|
|
|
| |
Two cleanups round out the transcription-config refactor:
- cj/--running-transcriptions: the 'status = running' filter used by
cleanup and count helpers is now one function. Existing counter tests
cover both callers.
- cj/--format-transcription-entry: the 13-line dolist body inside
cj/transcriptions-buffer becomes a testable pure function. 6 tests
cover status-face mapping, basename-only rendering, duration format,
trailing newline.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Break cj/--transcription-sentinel's seven inline side-effects into named
helpers:
- cj/--write-transcript-on-success: writes process output to .txt on success
- cj/--append-to-log: appends event marker + process output to log
- cj/--update-transcription-status: marks tracking-list entry complete/error
- cj/--notify-completion: sends success or critical notification
Also: switch the tautological (cj/--should-keep-log t) to use the local
success-p (equivalent but matches the function signature), and rename
the unused audio-file sentinel arg to _audio-file.
Sentinel shrinks from 48 lines with 7 inline blocks to 14 lines of
straight-line helper calls. 10 tests cover the extracted helpers.
|
| |
|
|
|
|
|
|
|
|
|
| |
Pull two more helpers out of cj/--start-transcription-process:
- cj/--init-log-file: writes the initial log header with timestamp,
backend, audio file, script path
- cj/--track-transcription: pushes a running-status entry and refreshes
the modeline
Start-process shrinks from 58 lines with 4 levels of nesting to ~25 lines
mostly at depth 1-2. 10 tests cover the extracted helpers.
|
| |
|
|
|
|
| |
Pull the per-backend env-var assembly out of cj/--start-transcription-process
into a standalone pure function. 9 tests cover: the three backends, parent-env
preservation, non-mutation, missing-key user-error, unknown-backend error.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
Introduce cj/--transcription-backends alist mapping each backend to
(:script :auth-host :env-var). Replace:
- two near-identical cj/--get-{openai,assemblyai}-api-key functions with
a single parameterized cj/--auth-source-password helper
- the pcase in cj/--transcription-script-path with an alist lookup
- the pcase block in cj/--start-transcription-process that assembled
the API-key env var with an alist-driven assembly
Adding a new backend is now a single line in the alist. The existing
tests plus retargeted API-key tests (now 10, covering the parameterized
helper and the descriptor data) verify no behavior change.
|
| |
|
|
|
| |
Dirvish uses its own keymap rather than inheriting from dired-mode-map,
so the transcription keybinding needs to be explicitly added.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Phone calls were not capturing the remote person's voice due to severe
volume loss (44 dB) when using the amerge+pan FFmpeg filter combination.
Changes:
- Replace amerge+pan with amix filter (provides 44 dB volume improvement)
- Increase default system volume from 0.5 to 2.0 for better capture levels
- Add diagnostic tool to show active audio playback (C-; r w)
- Add integration test with real voice recording
- Fix batch mode compatibility for test execution
The amix filter properly mixes microphone and system monitor inputs without
the massive volume loss that amerge+pan caused. Verified with automated
integration test showing perfect transcription of test audio.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This squash merge combines 4 commits from the performance branch:
## Performance Improvements
- **org-agenda cache**: Cache org-agenda file list to reduce rebuild time
- Eliminates redundant file system scans on each agenda view
- Added tests for cache invalidation and updates
- **org-refile cache**: Optimize org-refile target building (15-20s → instant)
- Cache eliminates bottleneck when capturing tasks
## Test Suite Improvements
- Fixed all 18 failing tests → 0 failures (107 test files passing)
- Deleted 9 orphaned test files (filesystem lib, dwim-shell-security, org-gcal-mock)
- Fixed missing dependencies (cj/custom-keymap, user-constants)
- Fixed duplicate test definitions and wrong variable names
- Adjusted benchmark timing thresholds for environment variance
- Added comprehensive tests for org-agenda cache functionality
## Documentation & Organization
- **todo.org recovery**: Restored 1,176 lines lost in truncation
- Recovered Methods 4, 5, 6 + Resolved + Inbox sections
- Removed 3 duplicate TODO entries
- **Inbox zero**: Triaged 12 inbox items → 0 items
- Completed: 3 tasks marked DONE (tests, transcription)
- Relocated: 4 tasks to appropriate V2MOM Methods
- Deleted: 4 duplicates/vague tasks
- Merged: 1 task as subtask
## Files Changed
- 58 files changed, 29,316 insertions(+), 2,104 deletions(-)
- Tests: All 107 test files passing
- Codebase: Cleaner, better organized, fully tested
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Integrated AssemblyAI as the third transcription backend alongside OpenAI
API and local-whisper, now set as the default due to superior speaker
diarization capabilities (up to 50 speakers).
New Features:
- AssemblyAI backend with automatic speaker labeling
- Backend switching UI via C-; T b (completing-read interface)
- Universal speech model supporting 99 languages
- API key management through auth-source/authinfo.gpg
Implementation:
- Created scripts/assemblyai-transcribe (upload → poll → format workflow)
- Updated transcription-config.el with multi-backend support
- Added cj/--get-assemblyai-api-key for secure credential retrieval
- Refactored process environment handling from if to pcase
- Added cj/transcription-switch-backend interactive command
Testing:
- Created test-transcription-config--transcription-script-path.el
- 5 unit tests covering all 3 backends (100% passing)
- Followed quality-engineer.org guidelines (test pure functions only)
- Investigated 18 test failures: documented cleanup in todo.org
Files Modified:
- modules/transcription-config.el - Multi-backend support and UI
- scripts/assemblyai-transcribe - NEW: AssemblyAI integration script
- tests/test-transcription-config--transcription-script-path.el - NEW
- todo.org - Added test cleanup task (Method 3, priority C)
- docs/NOTES.org - Comprehensive session notes added
Successfully tested with 33KB and 4.1MB audio files (3s and 9s processing).
|
| |
|
|
|
|
|
|
|
| |
Changed transcription submenu keybindings:
- C-; t t → C-; t a (transcribe audio)
- C-; t b → C-; t v (view transcriptions)
- C-; t k → unchanged (kill transcription)
More intuitive mnemonics: a=audio, v=view, k=kill
|
|
|
Implemented full transcription system with local Whisper and OpenAI API
support. Includes comprehensive test suite (60 tests) and reorganized
keybindings for better discoverability.
Features:
- Async transcription (non-blocking workflow)
- Desktop notifications (started/complete/error)
- Output: audio.txt (transcript) + audio.log (process logs)
- Modeline integration showing active transcription count
- Dired integration (press T on audio files)
- Process management and tracking
Scripts:
- install-whisper.sh: Install Whisper via AUR or pip
- uninstall-whisper.sh: Clean removal with cache cleanup
- local-whisper: Offline transcription using installed Whisper
- oai-transcribe: Cloud transcription via OpenAI API
Tests (60 passing):
- Audio file detection (16 tests)
- Path generation logic (11 tests)
- Log cleanup behavior (5 tests)
- Duration formatting (9 tests)
- Active counter & modeline (11 tests)
- Integration workflows (8 tests)
Keybindings:
- Reorganized gcal to C-; g submenu (s/t/r/c)
- Added C-; t transcription submenu (t/b/k)
- Dired: T to transcribe file at point
|
