Getting started
Overview
Toluva turns one approved source video into a time-aligned, consent-aware, verifiable localized edition — and keeps the evidence for how it did it.
Most tools will happily generate a translated voice track. Very few can tell you afterwards which voice was authorized for which language, how far each generated segment drifted from its source slot, what was rejected before it reached a provider, or which exact bytes were produced by which run.
Toluva is built around that second question. Backblaze B2 is the system of record rather than a place finished files are dumped, and every stage writes what it did before the next stage runs.
What it does
Consent-bound voice
Language, purpose, validity window and the consent-evidence hash are all checked before a provider is called.
Measured timing
Every translated segment is measured against its source slot, and drift outside the band triggers a bounded correction loop.
Verifiable lineage
Each stage is a separate run with a canonical manifest, and stored hashes are re-checked before bytes are described as verified.
What it does not claim
Toluva is deliberately narrow, and the interface is written to stay inside that boundary.
Scope
Toluva does not guarantee legal or regulatory compliance, does not support every language, does not produce perfect lip sync, and does not remove the need for human review. It is compliance-supporting and evidence-ready, which is a smaller and more defensible claim.
A canonical manifest proves recorded lineage and canonical integrity. It does not prove that every fact supplied to it was true.
The shape of a run
Intake
One short English clip, with source rights and synthetic-voice disclosure confirmed before anything is written.
Transcription and translation
The clip is segmented and timed, then translated with protected terms preserved.
Authorization
The requested language and purpose are evaluated against the stored authorization record. A request outside scope is refused here, before spend.
Speech and timing QA
Speech is generated per segment and measured. Segments inside the band are padded or tempo-fitted; segments outside it block and wait for approval.
Composition
Audio fan-in, captions, and a final render — each with its own manifest and hash.