Developed under the Horizon 2020 Programme, COMPRISE aims to define a fully private-by-design methodology and tools to reduce development costs and increase voice interfaces inclusiveness.

COMPRISE is of interest to all companies or administrations willing to deploy their own speech and language technologies solution or to store speech and language data in a GDPR-compliant way.

The project has several goals:

  • allow virtually unlimited collection of real-life speech and language data by anonymising it through innovative data transformations in such a way that the transformed data can be used to train speech-to-text (STT) and spoken language understanding (SLU) systems;

  • allow citizens to transparently access contents and services available in other languages by voice interaction in their own language;

  • reduce the costs for both technology providers and users of voice interaction technologies.

COMPRISE has reached many milestones since 2018! The clearest example is the progress made with the components that make up the solution:


  • The COMPRISE SDK, designed for smartphone applications, is developed using the Ionic framework on the Angular platform, allowing developers to create multilingual and voice-enabled applications in a faster, cost-effective, and privacy-driven way.

  • The COMPRISE Cloud Platform collects anonymised speech data to train STT and SLU models, and provides access to the resulting models.

  • The COMPRISE Voice Transformer converts each person’s voice into another’s and ensures that any information extracted from the transformed voice cannot be traced back to the original speaker.

  • The COMPRISE Text Transformer allows users in various application domains to mask critical information in a text to preserve privacy.

  • COMPRISE Weakly Supervised STT and Weakly Supervised NLU reduce the need for manual data labelling, a task both costly and time-consuming, when training STT and NLU systems.

  • COMPRISE Speech-to-Text Translation combines STT and Machine Translation to allow every user to speak their own language when interacting with a dialogue system that internally uses a different language.

Keep up with COMPRISE milestones at