On Finding Objects in Spectrograms: A Multiscale Relaxation Labelling Approach
This paper describes a new technique for finding objects in spectrograms, and illustrates the idea with an application to the formant-tracking task. Starting with a multiscale representation of speech spectra, a probabilistic relaxation labelling algorithm is applied to determine primitive interpretations of the spectral components. Finally, a cross-scale integration procedure enables the scale space to be collapsed in a principled manner. The techniques are illustrated with an example of voiced speech.
KeywordsEdge Weight Scale Space Label Probability Coarse Scale Coarse Level
Unable to display preview. Download preview PDF.
- /GRE 87/.P D Green, M P Cooke, H H Lafferty & A J H Simons: A Speech Recognition Strategy Based on Making Acoustic Evidence and Phonetic Knowledge Explicit, these proceedings (1987).Google Scholar
- /LEU 86/.H C Leung, V W Zue: Visual Characterisation of Speech Spectrograms. Proc. ICASSP, paper 51.1, (1986).Google Scholar
- /MAR 82/.D Marr: Vision, W H Freeman & Co, (1982).Google Scholar
- /WIT 83/.A P Witkin: Scale-Space filtering. Proc. IJCAI, (1983), 1019–1022Google Scholar