Topologically equivalent reconstruction of instationary, voiced speech

Voiced speech is characterized by qualitatively rich mode locking phenomena linking harmonically excited acoustic modes of the vocal tract. Due to the strong instationarity of speech, a differentiated analysis of these modes cannot be achieved with the help of a linear, time invariant source and filter model (based on stationary sources). As alternative, the characteristic mode locking is described as generalized synchronization in drive response systems with an instationary, common (fundamental) drive. By introducing a combined harmonic and logarithmic (audiological) scale subband decomposition adapted to the frequency of the master oscillator of phonation, a selfconsistently confirmed, topologically equivalent reconstruction of a number of acoustic modes of an acoustic object is generated. Whereas the invariant resonator properties (Lyapunov exponents) of the reconstructed response dynamics are characteristic for vowels, the generalized synchronization manifolds (lines or surfaces) in the combined state space of drive and respective response band can be used for the distinction of consonants. The topologically equivalent reconstruction of the phonation process is potentially useful for phoniatric diagnoses.