• Company
  Introduction and Overview Board of Directors
• Technology
  General Cognitive Electronic System References
• Business
  Consulting services Cognitive electronics
• Research
  Research Areas Research Projects
• Partners
• News
• Contact

Projects

Cognicatus together with div. of Electronics Systems, Dep't E.E. (Linköping University) currently run some final-year projects for Masters students in the fields of phoneme recognition and auditory scene analysis.

The algorithms are initially implemented and verified in software on a linux platform as open source projects. Then they will be implemented in a hardware platform, or hardware-in-the-loop kind of fashion, which targets ultra low power applications.

Applications in which the algorithms are used are hearing-aid devices, conference telephones, and mobile phones. In general, in applications in which we want to classify sound and potentially suppress sound that is not originating from a human speaker.

Implementation of a Phoneme Recognition Algorithm

To improve quality of sound we want to implement an algorithm that identifies phonemes in spoken language. This will enable the audio processing system to take more just decisions about the quality of sound, noise in channels, etc. Initially the algorithm will be able to only detect a limited number of phonemes in a “noise-free“ environment. Gradually we want to introduce distortion, noise, etc., to the channel and eventually define a software package in which we can adjust crucial parameters.

The objective of the algorithm is to determine and inform any observer which phoneme has been detected. This can be done either through a discrete set of value or through fuzzy logic (“this is probably a 't', but could be a 'd'”).

Implementation of an ASA Algorithm

The auditory scene analysis (ASA) field is large, but in our case we limit ourselves to investigate the quality of the sound. Firstly, this means that we will extract information of the sound: how much noise is there? What type of noise is it? Is it spoken language? Currently we are not looking deep into the field to further analyse sources of sound in the scene, spatial and temporal location, etc.

The objective of the algorithm is to determine and inform any observer which class of environment the microphone is. This can be done either through a discrete set of value or through fuzzy logic (“this is probably cocktail noise, this is probably wind noise, etc.).