Andreas Kromann
CEO Commercial lead
Andreas works with clients to turn model requirements into a concrete collection plan. Scope, speaker targets and delivery are agreed before production starts.
About us
A speech data company built around custom collection
We build custom speech datasets for voice AI teams training models across regional accents, multilingual speakers, and challenging everyday environments. From hard-to-source dialects to noisy background conditions, we help teams capture the exact speech patterns that off-the-shelf datasets consistently miss.
Why we exist
We remove the bottlenecks that stall speech data projects
Speech data projects break down in predictable places. Mismatched recruitment, fluctuating audio quality, fragmented vendor workflows, and rigid delivery structures consistently delay machine learning pipelines. We exist exclusively to eliminate these bottlenecks.
We focus entirely on sourcing hard-to-find speakers and managing the complete collection-to-delivery workflow under a single, rigorous quality standard.
In under a minute, Andreas explains the market problem we were funded by Innovation Fund Denmark to fix, and what we built instead.
What makes us different
We focus on speech data
A single continuous workflow built around the speakers, recordings, and formats your model actually needs.
Collection and transcription are handled within a single continuous workflow.
Recruitment matches your exact language and accent specifications.
Quality checks run actively alongside recordings instead of waiting for the final handoff.
Structured data packages integrate straight into your existing model format.
How we work
The operational philosophy
01
Specificity over scale
We deliver the precise linguistic data you need rather than overwhelming your team with bulk volume that misses the core specification.
02
Catch production issues live
Our internal reviewers sample active audio files during live recordings so that data faults are resolved before final delivery.
03
Single coordinated workflow
Collection, transcription, and validation happen within a single continuous chain to eliminate multi-vendor handoff gaps.
Meet the team
The team running your data collection project
We combine commercial project design, recruitment operations, platform engineering, transcription workflows, QA, and delivery management in one team.
Interested in contributing your voice or language skills? Learn about remote speech projects and register through the Spirelight contributor portal.
Gustav Aggeboe
CTO Platform architecture
Gustav leads the platform architecture. He builds the systems we use for recording, transcription, QA, metadata and dataset delivery.
Emil Thorsson
CFO Operations
Emil runs operations and compliance coordination. He keeps recruitment, consent, production and handoff in step from start to finish.
Joyi Ulfat
Senior Project Manager
Joyi manages execution across contributors, reviewers and delivery. She follows up on progress every day and makes sure each project meets its agreed requirements.
Mateo Thelen
Project Manager
Mateo coordinates contributors and recording workflows, turning project requirements into the daily work that keeps production moving.
Pekka Larjovuori
Crowd Source Expert
Pekka works on recruitment strategy and contributor operations, sourcing speakers for projects that need a specific language, dialect, region or profile.
Victor Melchior
Sales Market expansion
Victor leads sales and market research. He maps where our speech data fits new clients and regions, and opens the first conversations in each new market.
Yusif Aliyev
Legal Policy & compliance
Yusif leads legal, internal policy and compliance, keeping consent, data handling and contracts in order.
Michael J. Jørgensen
B2B partners & talent recruitment
Michael leads B2B sales and our partnerships with freelancers and companies, bringing in both the clients and the contributors our projects need.
Have a speech data project in mind?
Tell us the languages, accents, and conditions your model needs. We’ll scope the collection, recruit the right speakers, and deliver data in your format.
Book a call