This paper presents a real-time robust formant tracking system for speech signals and electroglottography (EGG) signals using a real-time phase equalization-based autoregressive exogenous model (RT-PEAR). PEAR can estimate formant frequencies robustly even for speech with high fundamental frequencies using phase equalization preprocessing and LPC with an impulse train. To reduce the computational complexity of original PEAR, a novel formulation of LPC with an impulse train is derived. EGG signals were used for stable detection of pitch marks since PEAR requires them. Formant estimation errors for the proposed method were less than 5 % regardless of fundamental frequencies with 12-ms processing delay. This technique will be useful for real-time speech conversion and speech-language therapy.

I invented a non-negative temporal decomposition method for line spectral pairs and articulatory parameters based on the multiplicative update rules. These parameters are decomposed into a set of temporally overlapped unimodal event functions restricted to the range [0,1] and corresponding event vectors. When line spectral pairs are used, event vectors preserve their ordering property. With the proposed method, the RMS error of the measured and reconstructed articulatory parameters is 0.21 mm and the spectral distance of the measured and reconstructed line spectral pairs parameters is 2.0 dB. The RMS error and spectral distance in the proposed method are smaller than those in conventional methods. This technique will be useful for many applications of speech coding and speech modification.

We present a non-negative temporal decomposition method for line spectrum pair and articulatory parameters. Based on the multiplicative update rules derived from a non-negative matrix factorization algorithm, these parameters decompose into a set of temporally overlapped event functions that are restricted to the range and corresponding event vectors. With the proposed method, the RMS error of the measured and estimated articulatory parameters is 0.16 mm and the spectral distance of the measured and estimated line spectrum pair parameters is 1.97 dB. These results also show that these estimation errors of proposed method are significantly smaller than those of the conventional method. This technique will be useful for many applications, such as speech coding and speech modification.

This paper presents a novel method for estimating a vocal-tract spectrum from speech signals, based on a modeling of excita-tion signals of voiced speech. A formulation of linear prediction coding with impulse train is derived and applied to the phase-equalized speech signals, which are converted from the original speech signals by phase equalization. Preliminary results show that the proposed method improves the robustness of the estima-tion of a vocal-tract spectrum and the quality of re-synthesized speech compared with the conventional method. This technique will be useful for speech coding, speech synthesis, and real-time speech conversion. Index Terms: LPC, vocal-tract spectrum, phase equalization 1.

We present a multi-speaker articulatory reconstruction method based on speaker-independent articulatory features and speaker-dependent features. These features are separated by using a multi-speaker articulatory database. This separation method consists of normalizing palate positions among the speakers and separating multi-speaker articulatory data into a speaker-independent eigen articulatory HMM and a speaker-adaptive matrix by using speaker-adaptive training (SAT). With the proposed method, the average RMS errors of the measured and reconstructed articulatory parameters were 1.35 mm. This result shows that the proposed method makes it possible to control speaker idiosyncrasies in the articulatory parameter domain.

We present a method that determines articulatory movements from speech acoustics using a Hidden Markov Model (HMM)-based speech production model. The model statistically generates speech spectrum and articulatory parameters from a given phonemic string. It consists of HMMs of articulatory parameters for each phoneme and an articulatory-to-acoustic mapping for each HMM state. For a given speech spectrum, maximum a posteriori estimation of the articulatory parameters of the statistical model is presented. The performance on sentences was evaluated by comparing the estimated articulatory parameters with the observed parameters. The average RMS errors of the estimated articulatory parameters were 1.50 mm from the speech acoustics and the phonemic information in an utterance and 1.73 mm from the speech acoustics only.

We present a method that determines articulatory movements from speech acoustics using an HMM (Hidden Markov Model)-based speech production model‥ The model statistically generates speech acoustics and articulatory movements from a given phonemic string. It consists of HMMs of articulately movements for each phoneme and an articulatory-to-acoustic mapping. For a given speech acoustics, maximum likelihood estimation of the articulatory parameters in the statistical model is presented. An experiment was conducted to evaluate the method on sentences, and estimated articulatory parameters were compared with the observed parameters. The rms error of the estimated articulatory parameters is about 2.12 mm on average.