Robust isolated speech recognition using ideal binary masks |
Seliz Karadogan, Jan Larsen, Michael Syskind Pedersen, Jesper B. Boldt
|
| Abstract | This is supplementary material for the paper of the same title.
The paper presents a new approach for robust ASR using ideal binary masks as feature vectors. This method is evaluated on a speaker-independent isolated digit database, TIDIGIT. Discrete Hidden Markov Model is used for the recognition and the observation vectors are quantized using K-means algorithm with Hamming distance. It is found that a recognition rate as high as 92% for clean speech is achievable using Ideal Binary Masks (IBM). It is also observed that IBM feature vectors are robust to different noise
conditions. |
| Keywords | isolated sppech recognition, robust, binary mask |
| Type | Technical report |
| Year | 2009 Month September |
| Publisher | Department of Informatics and Mathematical Modelling |
| Address | Richard Petersens Plads, Building 321, 2800 Kongens Lyngby, Denmark |
| Electronic version(s) | [pdf] |
| Publication link | http://www2.imm.dtu.dk/pubdb/p.php?5791 |
| BibTeX data | [bibtex] |
| IMM Group(s) | Computer Science & Engineering |