Cantab Research has expertise spanning 20 years in speech recognition, information retrieval, data mining, speech/music/image coding and many other areas related to human information processing.

Techniques used include statistical modelling and pattern recognition (HMMs, neural networks, Gaussian mixture models, vector quanitisation), machine learning (supervised, unsupervised, reinforcement, Bayesian inference), signal processing (FFT, LP, noise reduction), optimisation (EM, gradient descent, dynamic programming), search techniques (Viterbi, A*, time-first) and all aspects of speech analysis (MFCC, PLP), coding (LPC, CELP, MPEG), recognition, synthesis (concatentative, unit selection, PSOLA) and language modelling (n-gram models, backoff strategies).

Our speciality is applying these techniques outside of the areas that they were originally developed for.  For example to automatically generate themed playlists for iPod users we drew on advanced concepts in speech recognition.

Cantab Research can provide consultancy for rapidly emerging business oportunities.  We have an extensive library of algorithms and other resources to create, investigate and develop product ideas from proof of concept through to final product.