Characterizing resonant component in speech A different view of tracking fundamental frequency.pdfVIP

Mechanical Systems and Signal Processing 88 (2017) 318–333 Contents lists available at ScienceDirect Mechanical Systems and Signal Processing journal homepage: /locate/ymssp Characterizing resonant component in speech: A different view of tracking fundamental frequency Bin Dong Chair of Human-Machine Communication, Technical University of Munich, Theresienstra?e 90, 80333 München, Germany article info Article history: Received 27 August 2016 Received in revised form 17 November 2016 Accepted 19 November 2016 Available online 7 December 2016 Keywords: Speech signal processing Hilbert–Huang Transform Empirical mode decomposition Cyclostationarity analysis Modulation intensity distribution abstract Inspired by the nonlinearity and nonstationarity and the modulations in speech, Hilbert– Huang Transform and cyclostationarity analysis are employed to investigate the speech resonance in vowel in sequence. Cyclostationarity analysis is not directly manipulated on the target vowel, but on its intrinsic mode functions one by one. Thanks to the equivalence between the fundamental frequency in speech and the cyclic frequency in cyclostationarity analysis, the modulation intensity distributions of the intrinsic mode functions provide much information for the estimation of the fundamental frequency. To highlight the relationship between frequency and time, the pseudo-Hilbert spectrum is proposed to replace the Hilbert spectrum here. After contrasting the pseudo-Hilbert spectra of and the modulation intensity distributions of the intrinsic mode functions, it finds that there is usually one intrinsic mode function which works as the fundamental component of the vowel. Furthermore, the fundamental frequency of the vowel can be determined by tracing the pseudo-Hilbert spectrum of its fundamental component along the time axis. The later method is more robust to estimate the fundamental frequency, when meeting nonlinear components. Two vowels [a] and [i], picked up from a speech databa