Seeking Singapore ML partner: measuring spoken communication from speech

Most communication training is delivered as workshops with no way to assess where someone started, whether they improved, or how they progress over time. We close that gap with measurement.

Communication can be taught across four research-backed strands: physical, linguistic, cognitive, and social-emotional. This model is extensively validated in UK education as a way to teach communication. It has never been turned into a measurement instrument. We did something no one else had: we built a platform to measure communication against this validated model, not just teach it.

The first strand is live. It takes automatic speech recognition output, derives features such as word-level confidence, speech rate, and pause distribution, and maps them to a scoring model. The remaining three are specified but unbuilt and cannot be built yet. No annotated gold-standard corpus exists for these constructs in workplace speech, and construct validity has not been established. Building that corpus, training scoring models against it, and validating them against expert human judgement is the research programme.

The innovation is the measurement methodology, not off-the-shelf AI. Speech recognition can be bought, and an interface can be rebuilt. The corpus and the validated scoring models cannot.

We are looking for a Singapore-registered company with machine learning capability in speech, language, or multimodal data to lead model development and data engineering, while we lead construct design, annotation protocol, and validation. For the UK-Singapore CRD 2026 call, Digital and Technologies, AI. The Singapore partner applies in parallel to Enterprise Singapore.