Your recording choice
Updated September 17, 2026. SpeakEse is in private beta. Recording starts after you choose Start or take an eligible recording action from a notification. Recording hours set availability and reminders; they do not turn on the microphone by themselves. Get permission from other people where recording consent is required.
Normal app service
The phone detects speech and checks candidate audio clips. Approved conversation clips are uploaded for transcription and speaker separation. An approved clip can contain other people’s voices. My Voice uses the voice sample you agree to save. Final transcripts, audio positions, lessons, bookmarks, practice results and your choices are normal account records. Normal playback uses saved approved clips, so removed gaps and rejected audio are not included. Temporary local files support filtering, offline uploads and recovery. Verified uploads allow these local copies to be removed safely. Imported source files remain locally while needed to finish their conversations; normal imports do not upload the full source as a diagnostic archive.
Optional diagnostic sharing
During the friends pilot, debug tools are available by default. Extra sharing still needs your choice, and declining it does not block normal recording or learning. Extra diagnostics can include original conversation audio, accepted and rejected candidate clips, earlier transcription text and technical processing details. The founder can inspect this material in private Admin. Fast learning is separate and starts off. New diagnostic archives expire after 7 days; an outage can delay cleanup. Withdraw sharing in Account > Privacy & data > Diagnostic sharing to stop new extra collection on this phone and queue server deletion when online. Normal recordings and lessons remain. Older archives created under the earlier beta policy remain under that policy until deleted; the new feature does not silently delete historical recordings.
Working data and service measurements
The service keeps limited upload receipts, error codes, processing times and counts to deliver results, recover failures and understand reliability. Small lesson-exposure records help avoid counting an answer as unaided recall after an example was shown. These are separate from optional screen tracking. Temporary transcription checkpoints expire after 24 hours, lesson-exposure receipts after 30 days and the latest recording-health report after 24 hours; outages can delay cleanup. A requested lesson-audio timing check may send an approved clip to OpenAI to locate the original phrase. Its timing words never replace your transcript or speaker identity. Timing results remain with that recording. Exact phrase audio is shown only when the unchanged words can be located; otherwise the player explains that it is surrounding speech.
Practice and learning
OpenAI checks limited conversation excerpts and the practice text you submit. Spoken practice uses on-device speech recognition; its audio is not saved or uploaded. You can type instead and check the text before submitting it. Submitted answers are cleared after evaluation or by the pending-answer expiry worker after 15 minutes; outages can delay cleanup. Results and request receipts remain. You can pause learning, hide feedback types and choose reminders. Delete learning data removes lessons and progress without deleting recordings and pauses learning and reminders. Fast-test progress is kept separate from normal progress.
Optional Pilot feedback
Pilot surveys and detailed screen-action sharing have their own choice. They do not switch on just because debug access is enabled. Optional events use account-owned IDs, times, app-build details, an installation identifier and selected feedback, not audio, transcripts, practice answers, location or advertising tracking. Optional events and structured observations are kept for up to 90 days; an outage can delay cleanup. Stop sharing removes optional events and your survey answers when the request reaches the server. Existing service records and founder-entered support or quality observations are separate.
Private reports
Authorized administrators can download service and pilot reports with technical identifiers, counts and outcomes, and may share this metadata with an AI chatbot to understand product problems. These metadata reports contain no conversation text, audio or submitted practice answers and do not automatically contact an AI service. They are separate from consented audio diagnostic archives. Downloaded reports cannot be recalled automatically; contact support for help with exports already shared.
Access, providers and deletion
Amazon Web Services provides private storage, database and account services. OpenAI processes approved audio, the consented My Voice reference and learning requests. Authorized SpeakEse administrators can inspect records needed for support; enabling debug access does not give a customer Admin access. You can delete a conversation, remove My Voice or delete your account. Normal account records stay until deleted through these controls. Account deletion includes related product and diagnostic records. Provider-held data and backups follow their separate deletion processes. We do not sell information or use it for advertising. No security measure guarantees absolute protection. Contact support@speakese.com for privacy questions, a copy of your data or deletion help.