How can we help?

A quick guide to capturing your first conversation and getting back on track when something needs attention.

Share the right audio

Use a recent desktop Chrome or Edge browser. Choose a browser tab in the sharing picker and enable tab audio. Microphone-only mode uses a separate browser permission. Window and system audio vary by OS; mobile tab audio is often unavailable. Real meeting capture across browser/OS combinations still requires manual acceptance testing for this deployment.

Using a phone

Use the bottom tabs to start a conversation, open your library or change settings. If your browser supports microphone capture but cannot share meeting audio, microphone mode is selected for you. Review permissions, confirm consent and temporary local buffering, then start. Keep this page open and the screen awake; switching apps or locking your phone can interrupt recording. Other mobile apps’ meeting audio cannot be captured reliably. Try Check your audio and a short transcription first: local AI speed and memory use depend on your phone. Physical iOS and Android recording have not been verified for this deployment.

Test before an important meeting

To transcribe both yourself and the meeting, choose “Meeting + my voice”. Share the meeting tab with audio enabled, then allow microphone access. For laptop sounds outside a tab, select a window or screen only if the browser offers audio sharing. Wear headphones so meeting audio does not also enter your microphone. Both inputs are processed separately and labeled “Me” and “Meeting / laptop” in the saved transcript. These labels identify inputs, not individual people. Audio is never uploaded. You explicitly enable temporary local buffering before capture; processed audio is deleted. Separate processing increases the workload when both sources are speaking.

Open New transcription, choose your source, open “Check your audio”, and click “Test my audio”. Play speech or speak and watch the actual input meter. Stop the test, confirm participant consent, then start transcription. Audio testing does not download a model or save any content.

No words appearing?

Check that the audio meter moves. Initial model loading requires network access and may take several minutes. Listening starts once capture is ready and buffers audio while the model loads; recognition catches up once it is ready. Draft recognition starts with about 1.5 seconds of captured audio, plus processing time. Words can change as more context arrives; only finalized text is saved. A 5-second context window is a good starting point; longer windows give more context. The Detailed model is larger and slower; it may help recognition, but does not guarantee accuracy. Very quiet input is skipped to limit silence hallucinations. Draft updates slow down on busy devices so final text can take priority. Check the displayed queue and processing time. Use shorter sessions on slower devices.

A section is marked unrecognized

If the model gets stuck repeating text, KeepMyTalk retries that section locally with a different audio window before marking it unrecognized. Listening continues, and the warning updates when later speech is recognized. A retry can take extra processing time. Use headphones, check the selected microphone, and speak clearly. Unrecognized windows remain in this browser. Choose Retry processing, or recover the local session from New transcription. Successfully recognized audio is deleted; review important statements in the saved text.

Paused or overloaded?

Slow recognition does not pause capture. Pending audio is buffered on your device while the model catches up. A worker failure retries once, then offers Retry processing while listening continues. Pause capture only pauses when you choose it; that interval is not captured. Timestamps omit manual pauses. Local storage exhaustion, a revoked permission, disconnected device or an unrecoverable audio failure can still stop capture with an explanation. Switching devices mid-session is not supported; finish and start a new session.

Saving problems

Keep the session page open, reconnect, and retry saving. Export current text before leaving if saving remains unavailable. Unsaved completed text may be recoverable in this browser from the transcript detail page. Committed pending audio and checkpointed text can be recovered from New transcription in this browser and account. During a long finish, choose Finish later on this device to keep the remaining queue locally. The last partial window may be lost on an abrupt close. Signing out or deleting browser site data clears local recovery.

Getting into your account

Use “Create account” if you’re new or “Sign in” to return. Both continue securely through Google. Your account is created after your first successful Google sign-in; choosing Create account again returns you to your existing account. KeepMyTalk does not ask you to create a password.

If Google sign-in fails, reconnect to the internet and retry in a normal browser tab. Choose the same Google account you used before. If the page says sign-in is unavailable or not configured, the site owner needs to resolve the service setup. A Google authorization or redirect error also needs the site owner’s attention; retrying cannot fix incorrect OAuth settings.

Contact

Contact information has not been configured for this deployment. The source repository URL is also a deployment placeholder; no public GitHub link has been configured.