Our proprietary Phonologics pronunciation testing, called APST, is built upon knowledge-based speech analysis (KBSA), physics and acoustics, and the physiology of human speech production, modification, and enhancement
- pronunciation, accents, phonology
- signal processing and algorithm development
- English as a Second Language (ESL) training
- software development and technology
- human-machine interaction
Knowledge-Based Speech Analysis
KBSA is a system based on the methods and techniques of artificial intelligence. Two key components of the system are a knowledge base (derived from the expertise of a human “domain expert”) and inference mechanisms (a decision or classification engine).
The intellectual foundations for Phonologics‘ APST algorithms grew out of the manually-scored test platform developed over many years by Prof. Ferrier, a field-leading linguist. But APST radically innovates the model and methodology by which speech intelligibility has previously been determined.
Prof. Ferrier serves as the domain expert. Her expertise is used to determine which things to test for (e.g. the phonemes in a word like river) – as well as key features within those items that might have special significance (the first syllable, or letter combination, ri- ). APST uses KBSA to evaluate several components of intelligibility.
Easy, Accurate Testing
APST‘s advantage over expert listener methods is that it can freely use different words and phrases when testing non-native speakers of different first languages. This means no additional training is required to administer the test to a very wide variety of speakers. Within broad limits, the same flexibility allows the test to cycle through lists of suitably chosen words and sentences, using only a subset for any one speaker and sitting.
Research Funding from National Institutes of Health
The R&D phase of Phonologics‘ APST software was funded largely through National Institutes of Health (NIH) Small Business Innovation Research (SBIR) grants. Speech Technology and Applied Research Corp. (STAR) developed the technology. STAR’s expertise includes signal and image analysis and knowledge-based speech processing (KBSP), as well as statistical, mathematical, and system analysis. Most of the algorithm development in APST, particularly for KBSP and including the Phonologics scoring engine, uses Mathworks’ MATLAB quantitative computing environment. STAR is a Mathworks Connections Partner.