Thanks — this sounds frustrating, and I appreciate the clear details you shared. I’m responding to the two problems you described: (1)

createNoiseFilter

randomly throwing

WORKLET_NOT_SUPPORTED

mid-session and causing an infinite retry loop, and (2)

processor.restart()

throwing during reconnections so

sender.replaceTrack()

never runs and the mic stays dead. What I can confirm •

WORKLET_NOT_SUPPORTED

from

createNoiseFilter

can happen on Electron/Chromium desktop and may appear mid-session; when it happens the noise filter doesn't automatically recover and a full app restart is needed. • There is a defect where failed

createNoiseFilter

attempts can lead to an infinite retry loop: the hook sets

isNoiseFilterEnabled = false

on error, your code tries to re-enable it,

setNoiseFilterEnabled(true)

creates a new processor (the closure is stale on

krispProcessor

) and that new processor also fails — so it repeats and spams

WORKLET_NOT_SUPPORTED

/

DOMException: The user aborted a request

. • During room reconnections,

processor.restart()

may throw inside

setMediaStreamTrack

. Because there’s no

catch

(only a

finally

),

sender.replaceTrack()

does not run and the mic remains dead. Possible causes and what to try now • Worklet lifecycle / environment issue in Chromium/Electron: ◦ A worklet can become unusable mid-session if the underlying AudioWorkletGlobalScope or its supporting resources are torn down or blocked by the renderer process. Electron’s Chromium version, renderer restarts, or other native modules could cause this. ◦ Conflicting native integrations or other audio-related components in the app could interfere with the worklet environment. • Immediate mitigations you can implement client-side (safe to try now): a. Prevent infinite retries with backoff and caps • When

createNoiseFilter

fails, record the failure and do not immediately re-enable. Use exponential backoff plus a max retry count (for example: 1s, 2s, 4s, then stop and surface a user-visible error or a “Retry” button). • This prevents the loop and reduces error spam while you gather more info. a. Make processor lifecycle explicit and avoid relying on stale closures • Ensure the hook or code that calls

setNoiseFilterEnabled(true)

references the latest

krispProcessor

instance (e.g., keep the processor ref in a stable ref or state that updates, avoid

[]

deps for callbacks that need fresh references). • On failure, fully dispose the failed processor (stop tracks, disconnect any worklet ports) before attempting a new creation. a. Robust error handling around restart/replaceTrack • Wrap

processor.restart()

in try/catch. In the catch, attempt to recover by: ◦ disposing the processor, ◦ calling

sender.replaceTrack()

with the plain mic track (so audio still flows), ◦ then re-create the processor only when stable. • Always ensure

sender.replaceTrack()

runs in a finally or after you’ve handled thrown errors so the mic isn’t left dead. a. Add telemetry/logging • Capture timestamps, Chromium/Electron version, renderer process restarts, and exact error stacks when

WORKLET_NOT_SUPPORTED

appears. This helps correlate the failures with environment events. a. Fallback path for users • Provide a fallback “disable noise suppression” option that stops trying to create the processor and continues with raw mic audio until a manual retry. Information you can collect (these will help narrow root cause) • Exact Electron and Chromium versions where the issue reproduces. • Whether renderer process crashes/reloads or extension/native modules are present when the error happens. • Full error stack traces for

WORKLET_NOT_SUPPORTED

and

DOMException: The user aborted a request

. • Whether the problem ever resolves without a full app restart (you said it does not — confirm). • A minimal reproduction (if possible) that toggles the noise filter repeatedly and logs the worklet state. Short-term recommended plan 1. Add defensive handling now: ◦ Exponential backoff + max retries when

createNoiseFilter

fails. ◦ Catch around

processor.restart()

and guarantee

sender.replaceTrack()

runs to avoid dead mic. ◦ Make processor instance management explicit to avoid stale closures. 1. Collect the diagnostics above to help identify environmental triggers (Electron version, renderer events, other native modules). 2. If you can produce a minimal repro, share logs and versions so we can look for a pattern or a bug we can reproduce. If you’d like, I can: • Walk through a suggested code patch to add backoff + lifecycle fixes and show exactly where to add try/catch and processor disposal. • Help draft the logging/telemetry you should add to capture the needed fields. • Review a minimal repro or paste the relevant parts of your hook implementation so I can point out where the stale closure and missing catch should be changed. Which would you prefer to do next, and can you share the Electron/Chromium versions and the hook code that manages

krispProcessor

? Sources: [Flutter iOS] SoundWaveformWidget / AudioVisualizerWidget Not Working (Miss..., with interactive elements