Spire Light ApS

Spirelight was born out of a simple but persistent frustration in the voice AI industry: off-the-shelf speech datasets rarely match what production models actually need. Regional accents go uncovered, demographic representation is thin, and multi-vendor workflows introduce costly delays and quality gaps. The Copenhagen-based company was built specifically to solve these problems, offering a single continuous workflow that takes a client from initial data specification all the way through to a validated, ingestion-ready dataset.nnAt the core of Spirelight’s offering is its recruitment infrastructure. With more than 10,000 contributors in its network and active operations across 50-plus languages and dialects, the company can source native speakers matched to highly specific project criteria — by dialect, region, age, gender, device type, or acoustic environment. This granularity matters when training models that will be deployed in real-world settings, where a Jutlandic accent or a far-field kitchen recording environment can meaningfully affect model performance.nnThe company’s three main service pillars — speech collection, transcription and annotation, and dataset delivery — are designed to function as a single integrated workflow rather than separate hand-offs between vendors. Reviewers sample audio during live recording sessions, allowing quality issues to be caught and corrected before they propagate through the dataset. Clients can test early delivery batches and request mid-project adjustments without restarting production, a flexibility that distinguishes Spirelight from more rigid data suppliers.nnSpirelight has particular strength in Nordic languages and other underrepresented European language varieties, markets where off-the-shelf data is consistently sparse. The company holds SKI Verified Supplier status, making it accessible to Danish public-sector institutions, and carries a 5.0 rating on Datarade. Its client base spans automotive, smart home, wearables, call-centre technology, health tech, and any product category where natural human-machine voice interaction is critical.nnThe team is small but deliberately cross-functional, combining commercial project design, crowd-source recruitment expertise, platform engineering, and hands-on project management. This structure allows Spirelight to take ownership of the entire data lifecycle — from defining speaker targets with a client on day one to delivering a checksummed, manifest-packaged dataset to an S3 bucket at project close.