A sparsity based preprocessing for noise robust speech recognition
2014 (English)In: 2014 IEEE Workshop on Spoken Language Technology, SLT 2014 - Proceedings, 2014, 513-518 p.Conference paper (Refereed)
We show a method to sparsify the speech input that improves the robustness of an automatic speech recognizer. The proposed scheme is added to the system as a preprocessing module prior to the acoustic feature extraction. The preprocessing module passes the input speech signal through a linear predictive (LP) analysis filter and enforces sparsity in the LP residue domain. The sparsified prediction residue finally is filtered to generate the speech signal for computing a sequence of conventional feature vectors used in automatic speech recognition (ASR). Using standard feature vectors, our experiments show that sparsification in LP residue domain improves robustness in ASR performance.
Place, publisher, year, edition, pages
2014. 513-518 p.
Feature extraction, Linear predictive analysis, Residue signal, Robust speech recognition, Sparsity, Extraction, Speech, Speech communication, Acoustic feature extraction, Automatic speech recognition, Automatic speech recognizers, Noise robust speech recognition, Preprocessing modules, Speech recognition
IdentifiersURN: urn:nbn:se:kth:diva-167610DOI: 10.1109/SLT.2014.7078627ScopusID: 2-s2.0-84946689467ISBN: 9781479971299OAI: oai:DiVA.org:kth-167610DiVA: diva2:814775
2014 IEEE Workshop on Spoken Language Technology, SLT 2014, 7 December 2014 - 10 December 2014
QC 201505282015-05-282015-05-222015-05-28Bibliographically approved