Emotional Nuance

Speech Analysis

In the sound of speech there is a wealth of information about what the speaker was doing when they made the sound—including the movements of the mouth as it produced the sound, and the energetic state of the speaker, from which we can deduce facial expression. This is a type of “inverse problem”— reconstructing the cause (the speaker) from the observed effect (the sound).

As the first step of this inversion, our specialized speech analysis algorithms discover acoustic events in the audio, and trace those to back to appropriate kinds of movements in the mouth and face. Our algorithms work for any voice and are robust across recording conditions.

Making the Connection

When we started, the fields of speech technology and computer graphics were worlds apart. We built Carnival(TM) to bring them together: a modular fusion of speech recognition and other audio processing algorithms, human behavior modeling, physical modeling and 3D animation. This unified platform provides the backbone of all our software systems.

Our applications are built upon the core Carnival API and an ever-growing inventory of processing components embodying a wide range of algorithms. In its early days, Carnival, and its associated philosophy of software architecture, was featured in the September/October 2011 version of IEEE Computer Graphics and Applications. YES: it’s been around that long!

Continuous Improvement

As part of our commitment to deliver the highest quality facial animation at the fastest speeds, our software is subject to an ongoing programme of enhancements. Our roadmap is co-created with our clients, to ensure we focus on the things that matter most to you. New modules and updates can be adopted at a pace that suits you.