This paper investigates a new method for creating robust speaker models to cope with inter-session variation of a speaker in a continuous HMM-based speaker verification system. The new method estimates session-independent parameters by decomposing inter-session variations into two distinct parts: session-dependent and -independent. The parameters of the speaker models are estimated using the speaker adaptive training algorithm in conjunction with the equalization of session-dependent variation. The resultant models capture the session-independent speaker characteristics more reliably than the conventional models and their discriminative power improves accordingly. Moreover we have made our models more invariant to handset variations in a public switched telephone network (PSTN) by focusing on session-dependent variation and handset-dependent distortion separately. Text-independent speech data recorded by 20 speakers in seven sessions over 16 months was used to evaluate the new approach. The proposed method reduces the error rate by 15% relatively. When compared with the popular cepstral mean normalization, the error rate is reduced by 24% relatively when the speaker models were recreated using speech data recorded in four or more sessions.

This paper proposes a robust method for estimating the fundamental frequency (F0) in real environments. It is assumed that the spectral structure of real environmental noise varies momentarily and its energy does not distribute evenly in the time-frequency domain. Therefore, segmenting a spectrogram of speech mixed with environmental noise into narrow time-frequency regions will produce low-noise regions in which the signal-to-noise ratio is high. The proposed method estimates F0 from the periodic and harmonic features that are clearly observed in the low-noise regions. It first uses two kinds of spectrogram, one with high frequency resolution and another with high temporal resolution, to represent the periodic and harmonic features corresponding to F0. Next, the method segments these two kinds of feature plane into narrow time-frequency regions, and calculates the probability function of F0 for each region. It then utilizes the entropy of the probability function as weight to emphasize the probability function in the low-noise region and to enhance noise robustness. Finally, the probability functions are grouped in each time, and F0 is obtained as the frequency with the highest probability of the function. The experimental results showed that, in comparison with other approaches such as the cepstrum method and the autocorrelation method, the developed method can more robustly estimate F0s from speech in the presence of band-limited noise and car noise.