Refine
Document Type
- Diploma Thesis (1)
- Preprint (1)
Language
- English (2) (remove)
Has Fulltext
- yes (2)
Keywords
- Alter (1)
- Automatische Spracherkennung (1)
- Behinderter (1)
- Semantik (1)
- Signalanalyse (1)
- best basis (1)
- biorthogonal bases of L^2 (1)
- entropy (1)
- formants (1)
- local trigonometric packets (1)
Faculty / Organisational entity
We present results and views about a project in assisted living. The scenario is a room in which an elderly and/or disabled person lives who is not able to perform certain actions due to restricted mobility. We enable the person to express commands verbally that will then be executed automatically. There are several severe problems involved that complicate the situation. The person may utter the command in a rather unexpected way, the person makes an error or the action cannot be performed due to several reasons. In our approach we present an architecture with three components: The recognition component that contains novel features in the signal processing, the analysis component that logically analyzes the command, and the execution component that performs the action automatically. All three components communicate with each other.
* naive examples which show drawbacks of discrete wavelet transform and windowed Fourier transform; * adaptive partition (with a 'best basis' approach) of speech-like signals by means of local trigonometric bases with orthonormal windows. * extraction of formant-like features from the cosine transform; * further proceedingings for classification of vowels or voiced speech are suggested at the end.