Process

From buyer requirements to usable human data.

Sonexis connects buyer requirements, controlled contributor workflows, QA review, metadata, and delivery into one clean data process.

The process

Ten steps from scope to delivery.

Every project follows this sequence. The scope, volume, and timeline vary by buyer requirements, but the structure remains consistent.

1

Buyer shares requirements

Model type, use case, language needs, deployment context, and expected volume.

2

Sonexis scopes the dataset

Language combinations, speaker profiles, scenario types, metadata structure, and delivery format.

3

Collection plan is created

Task design, consent terms, QA standards, and timeline agreed before collection begins.

4

Contributors are matched through Collect

Screened contributors matched to language, region, and scenario requirements.

5

Tasks are assigned with clear guidelines

Scenario, format, recording environment, and consent terms included in every task.

6

Recordings are submitted

Contributors submit through Sonexis Collect with task metadata attached.

7

Submissions reviewed through QA

Audio quality, language match, consent verification, task accuracy, completeness.

8

Metadata and structure prepared

Speaker data, language tags, scenario notes, QA status, and known issues compiled.

9

Dataset is delivered

Structured delivery in the agreed format: audio, transcripts, metadata, manifests.

10

Feedback improves future batches

Model performance review informs the next collection cycle.

What buyers should prepare

What helps us scope accurately.

The more context you share, the more precisely we can scope the collection. You do not need all of this at the start, but the more detail you have, the faster we can move.

Model type
Use case and deployment environment
Language mix
Speaker requirements
Geography or region
Expected volume
Recording format
Transcript requirement
Metadata needs
QA standard
Delivery timeline
Compliance requirements
Delivery options

Six ways to work with Sonexis.

One-off custom dataset

A single scoped collection delivered against agreed requirements.

Evaluation dataset

Data structured for model evaluation against defined performance criteria.

Benchmark dataset

Standardised evaluation sets for repeatable cross-model benchmarking.

Ongoing collection pipeline

Repeatable, scheduled collection for teams with continuous data requirements.

Paid sample pack

A small structured sample to evaluate data quality before committing to a full collection.

Managed collection programme

Fully managed data collection with Sonexis owning design, collection, QA, and delivery.

Where Sonexis Collect fits

Collect is the operational layer.

Sonexis Collect is the platform behind contributor onboarding, consent, task assignment, recording submission, QA status, and contributor payment tracking. It is not a public upload service. Contributors are screened, tasks are assigned, and submissions are reviewed before any data enters a project delivery batch.

Start a dataset request.

Tell us what you are building. We will scope the collection and come back with a clear proposal.