Detection of acoustic phonetic landmarks is useful for a variety of speech processing applications such as automatic speech recognition.The majority of existing methods use Mel-frequency Cepstral Coefficients (MFCCs) describing the short time power spectral envelope of the speech signal. This paper hypothesizes that a different feature extraction method can be used to complement MFCCs by capturing more complex transient acoustic cues. The proposed feature extraction method quantizes spectrogram textures using local binary patterns (LBP). This paper particularly exploits landmark based stop consonant detection. Both methods outperform the previous work on stop consonant detection and the latter is particularly appealing for real time detection in which computation efficiency matters.
Cite as: Qian, K., Zhang, Y., Hasegawa-Johnson, M. (2016) Application of local binary patterns for SVM-based stop consonant detection. Proc. Speech Prosody 2016, 1114-1118, doi: 10.21437/SpeechProsody.2016-229
@inproceedings{qian16_speechprosody,
author={Kaizhi Qian and Yang Zhang and Mark Hasegawa-Johnson},
title={{Application of local binary patterns for SVM-based stop consonant detection}},
year=2016,
booktitle={Proc. Speech Prosody 2016},
pages={1114--1118},
doi={10.21437/SpeechProsody.2016-229}
}