Menu

Datasets

Labeled datasets for training and evaluation

Ezako creates and supplies high-quality labeled datasets for complex signals: acoustic, DAS, radio-frequency and time-series data. Datasets are labeled by domain specialists, structured for training and evaluation, and delivered as immutable versions so every model can be traced back to the data used to train and test it.

Supply

  • Off-the-shelf datasets

    Datasets built and labeled by Ezako, supplied under licence and usable for training on delivery.

  • Datasets to specification

    Ezako builds datasets from your recordings to your requirements, from annotation through to the delivered version. Work can be carried out by Ezako’s specialists at your premises or at Ezako. Alternatively, your team can build and manage the dataset directly in UpalgoDB on your own infrastructure.

Characteristics

  • Immutable versions

    Each version is frozen when it is built. A new version never alters a previous one, which keeps every trained model traceable to its exact data.

  • Partitioning

    Training, validation and test sets can be stratified or split by blocks, with separate hold-out sets where required. Data can be rebalanced and selected by source, filter, duration or other criteria.

  • Scenes

    Detections close in time are grouped into scenes, reviewed, retained or rejected scene by scene, and exported together.

  • Formats

    WebDataset and Hugging Face AudioDataset, with FLAC audio up to 96 kHz and labels in JSONL and CSV.

Public datasets used for demonstrations

What this site shows of Upalgo runs on public data, cited with its source: hydrophone recordings of the SanctSound project (NOAA), the C-MAPSS turbofan simulations of NASA (Saxena, Goebel, Simon and Eklund, IEEE PHM, 2008; 21 sensors per engine, four fleets run to failure), and a distributed acoustic sensing acquisition in the Trondheimsfjord (Thiem et al., Tracking ships with submarine cables, J. Acoust. Soc. Am. 159(5), 2026, CC BY 4.0). Customer datasets are never shown.

See what Ezako can do with your data

Describe your signals and requirements to discuss a dataset, labeling project or AI development workflow with our team.

Talk to the team