Speech & audio

Prompted, conversational, paired, and environmental audio with isolated tracks and transcripts.

Request a sample plan Review schema

Representative use cases

Speech recognition

Capture and acceptance criteria are tailored to the intended model behavior and deployment environment.

Conversational agents

Capture and acceptance criteria are tailored to the intended model behavior and deployment environment.

Acoustic event models

Capture and acceptance criteria are tailored to the intended model behavior and deployment environment.

Schema and capture coverage

Built for measurable coverage.

Every engagement begins with a versioned brief: intended use, environments, exclusions, participant consent, sensor configuration, quotas, and acceptance criteria.

media.uriProtected object reference · 48 kHz WAV · locale metadata · SNR checks

capture.contextLocale · coarse geography · environment · device

annotation[]Events · objects · actions · quality flags

rightsConsent version · allowed use · retention state

qaProbe version · reviewer agreement · disposition

deliveryManifest · checksums · limitations report

QA report preview · illustrative

Acceptance evidence travels with the data.

Coverage is reported against the approved quota matrix. Automated checks and blinded human samples expose rejection reasons, duplicates, media compliance, and annotation agreement.

Media probe pass98.4% illustrative

Human sample10% stratified

Agreement0.91 illustrative

Unresolved exceptions0 at delivery gate

From pilot to delivery.

Approve representative pilot samples before production. Inspect aggregate coverage, rejection reasons, agreement, duplicates, and media compliance as the collection progresses.

PILOT 01Environment coverage

PILOT 02Annotation overlay

PILOT 03QA evidence

Plan this dataset