ROBUSTNESS OF PHONEME CLASSIFICATION USING GENERATIVE CLASSIFIERS: COMPARISON OF THE ACOUSTIC WAVEFORM AND PLP REPRESENTATIONS
Matthew Ager, Zoran Cvetkovic, Peter Sollich and Jibran Yousafzai
In: ICASSP2008, 30 Mar - 04 Apr 2008, Las Vegas.
The robustness of classiﬁcation of isolated phoneme segments using generative classiﬁers is investigated for the acoustic waveform and PLP speech representations. Probabilistic PCA is used to ﬁt a density to each phoneme class followed by maximum likelihood classiﬁcation. The results show that although PLP performs best in quiet conditions, as the SNR decreases below 0dB acoustic waveforms have a lower classiﬁcation error. This is the case even though the waveform classiﬁer is trained explicitly only on quiet data and is then modiﬁed by a simple transformation to account for the noise, whereas for PLP separate classiﬁers are trained for each noise condition. Even at −18dB SNR, multiclass performance of classiﬁcation from waveforms is still signiﬁcantly better than chance level. In addition the effect of time-alignment is tested and initial solution shown.