Joint Frequency Domain and Reconstructured Phase Space Derived Features for Speech Recognition

A novel method for speech recognition is presented, utilizing nonlinear/chaotic signal processing techniques to extract timedomain based, reconstructed phase space derived features. By exploiting the theoretical results derived in nonlinear dynamics, a distinct signal processing space called a reconstructed phase space can be generated where salient features (the natural distribution and trajectory of the attractor) can be extracted for speech recognition. To discover the discriminatory strength of these reconstructed phase space derived features, isolated phoneme classification experiments are executed using the TIMIT corpus and are compared to a baseline classifier that uses Mel frequency cepstral coefficient features (MFCCs). The results demonstrate that reconstructed phase space derived features contain substantial discriminatory power, and when the two feature sets are combined, improvement is made over the baseline. This result suggests that the features extracted using these nonlinear techniques contain different discriminatory information than the features extracted from linear approaches alone. Because they attack the speech recognition problem in a radically different manner, these reconstructed phase space derived features are an attractive research opportunity for improving speech recognition accuracy.