WP3 – Massive adaptation

What is Massive Adaptation?

Massive adaptation can be thought of as a process of specialisation: general-purpose models are adapted to the specific variables of recorded lectures to create specialised “in-domain” models. The variables considered for the transLectures models include speaker, topic and type of language used. In addition, many of the video lectures on our case study sites are accompanied by time-aligned slides, from which key terms can be extracted and used to guide the system towards a correct translation/transcription.

Objectives

The main goal of this work package is to demonstrate that acceptable transcriptions and translations can be obtained via this process of massive adaptation. Three models are being developed in transLectures: acoustic, language and translation models, each of which is tackled under a separate task.

Deliverables

You can find information and download links for the transLectures tools in our Tools section.

Milestones

  • Initial models and tools for massive adaptation (M6)
  • First improvement of models and tools for massive adaptation (M12)
  • Second improvement of models and tools for massive adaptation (M24)
  • Third improvement of models and tools for massive adaptation (M30)
  • Final models and tools for massive adaptation (M36)

Click here for the full list of transLectures public reports and deliverables published to date.