Voicebook turns the conversations in your home — ad-hoc or ongoing — into a private, searchable archive of how your family actually talks, all processed on your own machine. From that one pipeline come the things you'll actually open: a child's language journal, a searchable family memory, a Sunday digest for the grandparents.
Leave it running in the background or capture a moment on the spot. Either way, one local pipeline turns raw sound into a structured record of who said what, in which language, over time.
The pipeline is the product; these are what it becomes once your home's voice is captured and indexed.
An auto-generated bilingual diary of a child's language — code-switching and all. Watch Claire's Mandarin and English grow month by month, in her own words, without transcribing a thing by hand.
Ask "what did Claire say about the museum last Sunday?" and jump straight to the moment — audio queued, in context.
A one-page Sunday summary — bilingual, side by side — for grandparents overseas who miss the kid between visits.
Off-the-shelf transcription falls apart on overlapping speakers, code-switching, and a three-year-old in a noisy kitchen. So the hard parts were rebuilt around that.
whisper.cpp and local models run on your own Mac. Your family's voice is never uploaded, brokered, or turned into someone's training data.
A pitch-analysis layer tags who's speaking by voice, catching a child even in a noisy room — the moments generic transcription would miss.
Tuned for mid-sentence Mandarin–English code-switching and your household's own vocabulary — nicknames, inside jokes, character names.
Cleanup only classifies lines that were actually said — the model is never allowed to write new text, so it can't fabricate a moment that never happened. See the pipeline →
Open source, local-first, and built around one family's real problem before it was ever a product. If that's your household too, let's talk.
Book a conversation