AuraOne / Human Data / Voice AI

Speech data built around the conversations your AI must handle.

Agree the capture, language, consent and quality requirements; receive the accepted recordings or evaluation outputs.

Speech is recorded with consent, and the permission stays with each recording.

Discuss voice data Explore Human Data

Input: Permissioned speech, scenarios, prompts, and release criteria
Output: Reviewed audio, scorecards, regressions, and safety findings

Voice programs

Collect the speech. Then judge it.

Two jobs, one record. Collection gives you speech you are allowed to use. Evaluation tells you whether the model got it right.

Collection packet

Speech: Bilingual conversation set, program consent
Transcript: Checked against the recorded speech
Speaker context: Language, accent, and scenario
Rights: Consent record and permitted use

Illustrative packet: field names only, no audio.

01 · Collection

Licensed speech collection

Speech recorded with consent for a defined program scope. Every file carries its rights and its speaker context.

02 · Evaluation

One relationship, judged at a time.

A reviewer checks whether each stretch of sound matches its word. Choose a word to see its group.

Human evaluation. Reviewers score how it sounds and whether it listened. Against criteria you set.
Regression benchmarks. The same test set follows every checkpoint. You see what got worse.
Safety programs. Impersonation, cloning, and scam scenarios, scoped into the program's safety review.

Concept illustration — no recorded audio

Scripted words, drawn form. Not a transcript.

A drawn violet sound form in 4 groups, one for each word of the scripted phrase “checking the morning route”. The selected word’s group is highlighted and connected to it.

“morning” selected: segment group 3 of 4, example 1 of 3.

About audio. This page plays no audio. The drawing and its words are scripted. Program audio stays with its consent record and its permitted-use scope.

From one recording to a release you can defend.

Clean audio is the easy part. The rights and the reviewer are what make it usable.

Speech you are allowed to use.

Speaker consent and usage scope
Language, accent, and scenario
Model, voice, or agent revision under review

Record it. Then review it.

Transcription and audio-quality review
Naturalness, pronunciation, and task scoring
Safety scenarios, overrides, and adjudication

A result you can reuse.

Accepted speech data and data card
Evaluation scorecard and reviewer rationale
Regression bank entry and release recommendation

Coverage planning

Plan coverage around release risk.

Coverage is scoped by target market, model behavior, and known failure modes, not a generic language count. The source and the reviewer stay attached to every file.

Define coverage

Market: Which languages and regions the release serves
Conditions: Noisy rooms, devices, and interruptions
Behavior: Code-switching, turn-taking, known failures

Groups teams often plan for

English accents and regional variants
Spanish and Portuguese variants
Indian languages and code-switching
Arabic regional variants
East and Southeast Asian languages
European languages and localization

Planning examples. Availability is confirmed per program.

Evaluation workflow

Voice evaluation walkthrough

The recording, the criteria, and the decision stay connected. Accept, rework, block, or release, with the approved scope stated.

Source: Bilingual conversation set under licensing reviewConsent record and license scope attached
Revision: Realtime agent candidate 24Baseline candidate 23
Review: Naturalness, latency, and safetyVersioned rubric and reviewer rationale
Decision: Hold for accent regression repairOwner and rerun criteria named

What the work lets you decide.

What the work lets you decide..
Outcome	Work	What you receive	Program fit
Accept a speech dataset	Capture it, transcribe it, and package the approved subset.	Consent, license, speaker context, data card.	Best fit: defined languages, scenarios, and acceptance criteria.
Release a voice model	Compare the candidate with the accepted baseline across target coverage.	Scorecards, retained regressions, and reviewer rationale.	Regional and safety cohorts remain visible beside the aggregate result.
Operate a voice agent	Test whether it finishes the task, takes its turn, and follows policy.	Conversation traces, rubric scores, and escalation state.	Live monitoring and provider controls are scoped per program where required.