Cross-Recording Voice ID.
Sarah on Tuesday's iPhone recording is the same Sarah on Friday's iPad recording — and every recording teaches Bonfiyah her voice a little better. One identity per person, every recording, every device, that gets sharper the more you record. Meeting bots start over every session; your speaker memory compounds.
Private to your account. Never used to identify anyone outside your own library.
One identity per person — that's the unlock.
Most recording apps treat every meeting as an island. Sarah from Tuesday's call has nothing to do with Sarah from Friday's call as far as the app is concerned — even though they're the same person. That's why "what did Sarah promise me last week" doesn't work in those tools, and why a profile of how Sarah communicates over time can't exist.
Bonfiyah recognises Sarah across every recording she's in. One identity per person, regardless of which device captured her, when, or alongside whom. That single fact is what makes everything downstream — Promise Tracker, Speaker Insights, Pre-Brief, Team Dynamics — actually work.
It also handles the silent-error case where two people with similar voices get merged into a single label by the upstream transcription — one tap on Split by Voice separates them back into the right people by voiceprint.
What identity stability buys you.
Promise Tracker
"What did Sarah promise me last week" only works if Sarah is one identity, regardless of which device captured the recording. Voice ID is the precondition.
Speaker Insights · People Memory
A profile of every person you've recorded with is only coherent if "every recording with Sarah" is computable. Identity stability is what makes the profile real.
Compatibility Analysis
Pairwise frameworks (Attachment, Big Five, Gottman, Thomas-Kilmann) anchor to two specific identities. Mixed-up labels break this entirely; we won't run Compatibility on a recording with unresolved labels.
Team Dynamics
A team is a selection of identities, with placements anchored across many recordings. Identity stability is what makes the longitudinal read possible — watch Riley's placement migrate across a quarter only because Riley is one identity.
Privacy
Voice signatures are biometric data. We treat them that way.
Your voice signature is computed on your device. Recognition — matching that signature against your existing speakers — runs on our backend over a TLS-encrypted connection, using a purpose-built voice-recognition model tuned specifically for short, conversational utterances. The audio used to recognise a speaker is the audio of the recording you already authorised — nothing extra is captured, the fingerprint can't be reversed back into audio, and nothing is held longer than it takes to match.
Voice signatures are biometric identifiers under GDPR, BIPA, and most modern privacy regimes. We treat them that way. Yours are stored under your account, sent only over HTTPS, and isolated from every other Bonfiyah user's library — there is no shared index and no "who is this voice on the internet" pathway. Unused signatures auto-purge after 90 days of inactivity. Delete a speaker manually any time, and the fingerprint goes with them.
Voice signatures are not the recording. They are a numerical fingerprint of the voice, and the original audio cannot be reconstructed from them.
Delete a speaker from your library and the signature is purged. Delete a recording and the signatures derived from it are purged with it. Uninstall Bonfiyah and the recognition system goes with the app — there is no off-device residue.
✦ Self-healing
Speaker memory that cleans itself.
Most tools treat speaker labels as disposable — they re-identify everyone from scratch every meeting and reset at the next one. Bonfiyah does the opposite. Your speaker memory is durable, and it actively repairs itself: it automatically finds the contaminated voice data — a stray sample, a crossed wire between two similar voices, a profile that's quietly drifted — and reversibly removes it (quarantined, never deleted, restorable exactly), so each person's voiceprint stays clean as new conversations come in. It runs on its own, in the background, with nothing for you to manage.
The result is a profile that gets sharper and more accurate the more you use it, with nothing for you to manage. Every recording is more signal about how each person actually sounds — across rooms, microphones, and moods — so recognition compounds instead of starting over.
It's consent-bounded by design: voices that withheld consent are never used to train your speaker memory, and you can reassign or remove any identity at any time. The longer you're with Bonfiyah, the better it knows the people in your life — a head start a brand-new app can't copy.
Where Voice ID will mis-recognise.
We're explicit about the edge cases because anyone using Bonfiyah for legal, medical, or research workflows needs to know them.
Siblings and identical twins
Voices that share substantial acoustic structure can be bound to one identity. Tap any line in the transcript to reassign — the fix persists across the library and future recordings.
Voice age drift
A child's voice changes substantially over months. A speaker entry that hasn't been refreshed in a long while may not match the same person today. The library updates on every successful match, so this resolves on its own with regular use.
Phone calls vs. in-room
A speaker on a low-bitrate phone call sounds different to the same speaker captured in-room. Bonfiyah handles most channel mismatches, but extreme cases (a cellular call vs. studio-quality lavalier) can need a manual reassignment the first time.
Tap to fix anything wrong
Every transcript line is one tap away from a speaker picker — pick the right person and Bonfiyah carries the correction forward. No support ticket, no menu diving.
FAQ
Can Voice ID be used to identify someone outside my library?
No. There is no global voice database, no cross-account index, no "who is this voice on the internet" pathway. Identity is scoped to your private speaker library, which is scoped to your iCloud. Bonfiyah is a memory layer, not a surveillance layer.
Where does the recognition run?
The voiceprint is computed on your device; matching runs on our backend over a TLS-encrypted connection, using a purpose-built voice-recognition model. The audio used is the audio of the recording you already authorised — nothing additional is captured. Voice signatures are isolated to your account, never shared across users, and auto-purge after 90 days of inactivity.
What if Bonfiyah gets a speaker wrong?
Tap the line and pick the right person. The fix carries forward to every future recording, and any duplicate identities created by the mistake are merged. And when the upstream transcription merges two similar voices into one label, Split by Voice separates them back into the right people by voiceprint in a single tap. On top of that, Bonfiyah's speaker memory heals itself — it keeps checking each voiceprint against your recordings and quietly cleans up the occasional misheard speaker, so accuracy improves the more you use it.
Is BIPA-style consent required?
In jurisdictions where biometric-data handling requires informed consent (Illinois BIPA, similar regimes), Bonfiyah's consent flow includes the voice-recognition disclosure as part of standard recording consent. We're not lawyers and this isn't legal advice; the disclosure is in plain language so you and your counsel can confirm it covers your jurisdiction.
Can I export my speaker library?
Yes. The export contains named identities and per-recording statistics. The underlying voice signatures are biometric data and don't export by default — there's an explicit "include voice signatures" toggle for portability or compliance scenarios, with a clear warning about what you're choosing to export.
Stay near the fire
Bonfiyah notes — feature deep-dives, real-world recipes, the occasional model explainer like this one.
No spam. We use ConvertKit. See our privacy policy.