Jatin K MalikandGitHub
65de3427fb
* fix(perf): reduce startup and inference overhead * fix(perf): report effective whisper.cpp fallback settings * fix(injection): add opt-in X11 paste for long dictations * perf(vad): lower default silence timeout to 1.2s * fix(types): resolve mypy errors across application source * ci: run mypy in the lint job * ci: include mypy in lint dependency group * fix(types): narrow evdev timestamp comparison * refactor(recognition): add SpeechEngine protocol and engine dispatch table * refactor(settings): move About section and engine tables out of settings_dialog * docs: add ARCHITECTURE.md and refresh docs index and AGENTS layout * fix(settings): preserve GTK dialog compatibility after extraction * style: format touched Python modules * fix(types): keep local typecheck from syncing dependencies * test(vad): align main startup expectation with new default * style: sort engine registry and protocol test imports * test(settings): follow extracted About builder attribute alias * test(settings): isolate GTK dialog base in About test * test(settings): isolate About color scheme build test * test(settings): restore GTK base after About scheme test * test(settings): isolate About scheme check from GTK mocks * test(settings): support extracted dialog mixin methods * docs: add before/after performance benchmark charts * fix(injection): type instead of pasting when the clipboard holds non-text data
57 lines
2.3 KiB
Markdown
57 lines
2.3 KiB
Markdown
# Performance benchmarks
|
|
|
|
Before and after numbers for the performance pass in
|
|
[#968](https://github.com/VocaHQ/vocalinux/pull/968). "main" is the code before
|
|
that PR. All runs used the same box: 8 vCPU Intel Xeon Platinum 8559C, Python
|
|
3.13, pywhispercpp 1.5.1, X11.
|
|
|
|
## Settings first open
|
|
|
|
|
|
|
|
The Settings dialog used to import every installed engine package (Vosk,
|
|
OpenAI Whisper, faster-whisper) just to see whether it was installed. With all
|
|
optional engines installed, the first open took 1.38 s and the process kept an
|
|
extra 606 MB for the rest of its life. Probing with `importlib.util.find_spec`
|
|
brings that down to 214 ms and 50 MB.
|
|
|
|
## Text injection on X11
|
|
|
|
|
|
|
|
Measured by injecting text into KWrite through `TextInjector.inject_text`,
|
|
mean of 3 runs. Dropping the 0.1 s sleep after every 20 characters saves
|
|
about 2 s per 420 characters. Turning on Settings > "Paste long dictations"
|
|
sends text of 30 or more characters through the clipboard, which takes about
|
|
40 ms at any length; the previous clipboard content is restored afterwards.
|
|
Typing speed itself is set by xdotool's per-key delay, so absolute typed
|
|
times vary by machine.
|
|
|
|
## whisper.cpp after a GPU init failure
|
|
|
|
|
|
|
|
When Vulkan or CUDA fails to initialise, the CPU fallback used to keep the
|
|
GPU thread count (cores / 4, so 2 here) instead of the CPU policy
|
|
(cores - 2, capped at 8, so 6 here). Times are for an 11 s clip.
|
|
|
|
## Silence timeout
|
|
|
|
|
|
|
|
200 LibriSpeech dev-clean utterances (5 per speaker, 26 min of audio) were
|
|
replayed through the same segmentation `_record_audio` uses: 1024-sample
|
|
chunks, Silero VAD at the default sensitivity, and a split whenever silence
|
|
exceeds the timeout. Below 1.2 s some sentences get cut in two at a natural
|
|
pause; at 1.2 s none were. The default is now 1.2 s instead of 2.0 s, so each
|
|
dictation finishes 0.8 s sooner. Read speech pauses less than live dictation,
|
|
so raise the timeout in Settings if you pause mid-sentence to think.
|
|
|
|
## Where the time goes
|
|
|
|
On an 11 s utterance with the tiny model, the Python code around the
|
|
transcription took 0.3 ms of a 389 ms call; the rest is native whisper.cpp.
|
|
A standalone `whisper-cli` built from the same whisper.cpp commit was no
|
|
faster, so rewriting the app in another language would not speed up
|
|
transcription.
|