fix(settings): measured CPU and download figures for the noise-suppression models #481

Merged
jared merged 1 commits from fix/denoise-cpu-labels into lotus 2026-10-29 21:57:15 -04:00
Owner

The CPU figures in Settings → Calls were guesses with no unit, so they read like Task Manager's whole-CPU number and looked wildly off. I measured each model running exactly as Lotus Call runs it: real-time speech, browser-process CPU time from /proc minus a no-model baseline, on a Ryzen 7 PRO 4750GE, repeated twice.

Model Was Measured (one core) Download (fetched)
DeepFilterNet 3 25–50% ~30% 17.6 MB (label ~18 MB ✓)
DTLN 10–20% ~9% ~20 MB (was "~4 MB")
RNNoise < 5% ~3% 0.2 MB
Speex < 2% ~2% 0.1 MB
  • The CPU stat is now labelled "one core".
  • The model description explains that Task Manager shows this figure divided by your core count, and that slower computers use more.
  • lotusDenoiseUtils.ts records how the numbers were measured.

🤖 Generated with Claude Code

https://claude.ai/code/session_017PCb59uY3UKqDqt7B6hzd5

The CPU figures in Settings → Calls were guesses with no unit, so they read like Task Manager's whole-CPU number and looked wildly off. I measured each model running exactly as Lotus Call runs it: real-time speech, browser-process CPU time from `/proc` minus a no-model baseline, on a Ryzen 7 PRO 4750GE, repeated twice. | Model | Was | Measured (one core) | Download (fetched) | |---|---|---|---| | DeepFilterNet 3 | 25–50% | **~30%** | 17.6 MB (label ~18 MB ✓) | | DTLN | 10–20% | **~9%** | **~20 MB** (was "~4 MB") | | RNNoise | < 5% | **~3%** | 0.2 MB | | Speex | < 2% | **~2%** | 0.1 MB | - The CPU stat is now labelled "one core". - The model description explains that Task Manager shows this figure divided by your core count, and that slower computers use more. - `lotusDenoiseUtils.ts` records how the numbers were measured. 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_017PCb59uY3UKqDqt7B6hzd5
jared added 1 commit 2026-10-29 21:35:38 -04:00
fix(settings): measured CPU and download figures for the noise-suppression models
CI / Build & Quality Checks (push) Skipped
CI / Secret scan (gitleaks) (push) Skipped
CI / Playwright smoke (e2e) (push) Skipped
CI / Build & Quality Checks (pull_request) Successful in 1m40s
CI / Trigger Desktop Build (pull_request) Skipped
CI / Docker image build & smoke test (pull_request) Skipped
CI / Secret scan (gitleaks) (pull_request) Successful in 6s
CI / Playwright smoke (e2e) (pull_request) Successful in 19m45s
1ebc42863e
The CPU column was guessed (DeepFilterNet "25-50%", DTLN "10-20%") with no
unit, so it read like Task Manager's whole-CPU figure and looked wildly off.
Measured with each model running exactly as Lotus Call runs it (real-time
speech, browser process CPU time from /proc minus a no-model baseline,
Ryzen 7 PRO 4750GE): DeepFilterNet ~30%, DTLN ~9%, RNNoise ~3%, Speex ~2%
of one core. Label it "one core" and explain the Task Manager relation.
DTLN's download was listed as ~4 MB but fetches ~20 MB (17 MB worklet with
the runtime inlined, plus 4 MB of models); DFN's ~18 MB is right (17.6 MB).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017PCb59uY3UKqDqt7B6hzd5
jared scheduled this pull request to auto merge when all checks succeed 2026-10-29 21:35:45 -04:00
jared merged commit d2c3c56b18 into lotus 2026-10-29 21:57:15 -04:00
jared deleted branch fix/denoise-cpu-labels 2026-10-29 21:57:15 -04:00
Sign in to join this conversation.
No Reviewers
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: LotusGuild/cinny#481