Pro AI
Your speaker memory gets sharper every conversation — and heals itself.
Most apps start from zero on every recording. Bonfiyah doesn't. It remembers the people you record by the sound of their voice — and that memory compounds: the more you record, the more reliably it knows everyone. It even self-heals, quietly re-checking its own voiceprints and cleaning up any that drift. Speaker Memory Health is your window into that living memory — person by person, how well each voice is known, and the rare few worth a quick confirmation. All from voice biometrics, never from what was said.
It learns over time — and that's the whole point.
Every conversation you record adds a little more voice to each person's profile. A name that started as "still learning" off a few seconds of audio becomes rock-solid after a handful of meetings. Six months in, Bonfiyah recognizes your team, your family, your regulars the instant they speak — across recordings, across devices, without you tagging anyone. Your speaker memory is a durable, growing asset that belongs to you and gets more valuable the longer you use it.
That's something a meeting-bot or a one-shot transcriber simply doesn't have. They treat every recording as a blank slate. Bonfiyah carries who-said-what forward and sharpens it — the compounding advantage that makes month twelve far better than month one.
And it heals itself.
Voice recognition can drift. A noisy room, a cold, a label that accidentally captured two people who sound alike — left alone, small errors compound the wrong way. So Bonfiyah runs a continuous self-healing pass over your speaker memory: it re-scores each voiceprint against everything it has learned, spots the few that have drifted or blended two people together, and reversibly cleans them up — keeping your recognition accurate without you having to babysit it.
Speaker Memory Health is where you watch that happen: which voices are strong, which are still learning, and the rare one flagged as "may include more than one person" before it ever causes a mix-up. The engine does the work in the background; this is your clear, plain-language view of it staying clean.
Recognition that compounds
Your speaker memory gets sharper the more you record — and self-heals when a voiceprint drifts. Every other app starts from zero on each recording.
Illustrative — recognition quality is computed from voice biometrics only, never from what was said.
What each person's card shows
- 🟢 Recognition-quality band — Strong (a clear, well-separated voiceprint), Still learning (usable, but only a little voice to go on yet), or Needs attention (thin, ambiguous, or possibly mixed). The one-glance answer to "does Bonfiyah know this voice?"
- 🎙️ Voice samples learned — how many distinct stretches of this person's voice Bonfiyah has gathered. More samples, sharper recognition.
- 👥 "May include more than one person" flag — raised when a single voiceprint looks like it may actually hold two different voices, so you can split them apart before attribution drifts.
- 🔀 "Might be confused with someone" hint — raised when two people sound similar enough that recognition could occasionally cross them, with the name of the person it might be mixed up with.
How it's computed
Every reading comes from voice biometrics — the acoustic distinctness of each person's voiceprint and how much voice Bonfiyah has gathered for them. The bands, sample counts, and flags are all about the recognition: how separable one voiceprint is from another, and how well-supported each one is. None of it reads the transcript. What anyone said never factors into a recognition-quality reading.
It's a read-only view of your own account. Opening Speaker Memory Health doesn't change anything in your library — it just shows you the state of your recognition and, where a quick fix would help, makes it one tap to confirm or reassign.
Your voiceprints stay yours
Speaker Memory Health is your own window into your speaker recognition — built entirely from your own recordings, scoped to your account, and held to the same AI-processing consent and per-feature controls as every other Pro AI surface. Voices that withheld consent are never used to train recognition, and you can reassign or remove any identity at any time. It pairs naturally with Voice ID and People Memory: those build the recognition; this shows you how it's doing.
FAQ
What is Speaker Memory Health?
It's a private, per-person view of how well Bonfiyah currently recognizes each person's voice. For each person you've recorded, it shows a recognition-quality band — Strong, Still learning, or Needs attention — how many voice samples Bonfiyah has learned for them, whether their voiceprint may include more than one person, and whether it might be confused with someone else you record. It's a diagnostic view on top of Bonfiyah's voice recognition, not a new score about the person — it's about the quality of the recognition itself.
Does Bonfiyah's speaker memory get better over time?
Yes — that's the core of it. Every conversation adds more voice to each person's profile, so recognition sharpens the more you record: a voice that's 'still learning' off a few seconds becomes reliably 'strong' after a handful of meetings, and that memory carries forward across recordings and devices. Bonfiyah also self-heals — it continuously re-scores each voiceprint against everything it has learned and reversibly cleans up any that have drifted or accidentally blended two people together, so accuracy improves rather than decays as your library grows. Speaker Memory Health is your window into that: a durable, compounding voice memory that's worth more the longer you use it.
Does it analyze what people said?
No. Recognition quality is computed from voice biometrics only — the sound of each voice — never from the words in the transcript. The bands, sample counts, and flags all describe how distinct and well-learned a voiceprint is, not anything about the content of anyone's speech.
What do the recognition-quality bands mean?
Strong means Bonfiyah has a clear, well-separated voiceprint and recognizes this person reliably. Still learning means it has a usable voiceprint but only a little voice to go on yet — recognition will sharpen as you record more conversations with them. Needs attention flags a person whose voiceprint is thin, ambiguous, or possibly mixed with someone else, so you may want to confirm or reassign a recording to keep your history attributed correctly.
What are the 'more than one person' and 'might be confused with' flags?
When a single voiceprint may actually contain two different people's voices, Speaker Memory Health raises a 'may include more than one person' flag so you can split them apart. When two people sound similar enough that recognition could occasionally cross them, it raises a 'might be confused with <name>' hint. Both are early warnings that let you fix attribution before it drifts — your conversation history stays attached to the right people.
Are my voiceprints private?
Yes. Your voiceprints stay yours — Speaker Memory Health is simply your own window into them. It's built entirely from your own recordings, scoped to your account, and runs under the same AI-processing consent and per-feature controls as every other Pro AI surface. The self-healing happens only within your own account; voices that withheld consent are never used to train recognition, and you can reassign or remove any identity at any time.
A voice memory that gets sharper every conversation.
Speaker Memory Health is a Pro AI surface. We'll let you know as it rolls out, plus the occasional note on how Bonfiyah's voice recognition learns and heals over time.
No spam. We use ConvertKit. See our privacy policy.