BrainSpace's solutions are based on audio analysis technology developed by SoftSound, which uses advanced statistical methods to deal with all aspects of processing the digital audio signal. It employs a wide range of recognition technologies - from keyword and phrase spotting to small and large vocabulary continuous speech recognition, speaker recognition and language recognition. Because of the accuracy and utility of the speech recognition, SoftSound's technology has consistently outperformed other systems for the precision of its real-time anonymous speech recognition tasks as benchmarked by:
DARPA and NIST (both US Government agencies)
THISL (an EU research initiative)
SQUALE (a pan-European evaluation)
To achieve anything near the level of precision that Softsound's recognition engine achieves, the closest of the other systems required a massive increase (around ten times) in hardware investment and their associated running costs, to get close to the precision that the SoftSound technology delivers on a standard Pentium box.
Softsound's audio analysis technology uses neural network technology and "Hidden Markov Models" (HMMs) to construct a highly efficient acoustic model, which is able to provide a fast, accurate and dynamic solution within variable and rapidly changing acoustic environments. The system technology is based around decomposing digitized speech into its phonetic constructs. The phonetic sequence is then analyzed in conjunction with acoustic model and statistical improbabilities to calculate which is the most probable sequence of hence words and utterances. The major benefits of the approach to speech processing and recognition taken by the BrainSpace audio analysis technology are described below:
2002 - 2004
