Data processing & annotation, human-in-the-loop.
Stepe Digital SIA delivers Latvian-language transcription and audio annotation for language-modeling pipelines. Reviewed by people, documented end to end, and governed by EU law.
Annotation services built for training pipelines
Every deliverable is human-reviewed and shipped in the schema your pipeline expects.
Transcription & HITL correction
Latvian-language audio transcription with human review, word-level timestamps, and speaker labels.
Speaker diarization
Multi-speaker audio segmentation and annotation, structured for supervised training datasets.
Custom formatting
Deliverables in client-specified formats: CSV, JSON, or your own schema, ready for ingestion.
Quality assurance
Second-pass review against your acceptance criteria, with per-batch quality summaries.
A documented path from scope to delivery
Every engagement runs under a project-level agreement defining scope, timeline, and price.
Scope & DPA
We agree scope and sign a Data Processing Agreement before any data changes hands.
Annotate
Screened native-Latvian linguists transcribe and annotate to your documented instructions.
Review
A second pass verifies accuracy and records error rates for the batch.
Deliver & purge
Data ships in your schema; source material is deleted within 30 days of completion.
A registered EU data processor you can put in a contract
Stepe Digital SIA (Reg. No. 40203711274) is a registered Latvian company operating under EU law. We act as a GDPR-compliant Data Processor under Article 28, handling client audio data strictly on documented instructions. A Data Processing Agreement is available on request.
All engagements are governed by project-level agreements specifying scope, timelines, and pricing. Invoices are issued in EUR, and client data is permanently deleted within 30 days of project completion.
Have a dataset that needs human review?
Tell us the language, volume, and target schema. We'll scope the work and send a DPA within a few business days.