Existing Reference Models Research Articles

Adaptation of hidden Markov model for telephone speech recognition and speaker adaptation

The authors propose a channel compensation method for the hidden Markov model (HMM) parameters in automatic speech recognition. The proposed approach is to adapt the existing reference models to a new channel environment by using a small amount of adaptation data. The concept of HMM parameter adaptation by incorporating the corresponding phone-dependent channel compensation (PDCC) vectors is applied to improve the performance of speech recognition. Two extended PDCC techniques are presented. One is based on the refinement of PDCC using vector quantisation. The other is based on the interpolation of compensation vectors. Both techniques are evaluated on the experiments on telephone speech recognition and speaker adaptation. The experimental results show that the performance can be significantly improved.

Constraints on the shallow crustal model of the Northern Apennines (Italy) from the analysis of microearthquake seismic records

SUMMARY A joint study of microearthquake source and medium parameters was carried out by analysing a seismic sequence that occurred in the Tuscan Emilian Apenninic region. Signal processing and graphic techniques were applied to study the amplitude and polarization of wave motion and thus identify the main and secondary body-wave arrivals on the seismic records. The evidence for similar waveforms (denoted as 'similar events') in a microearthquake subset allowed application of a non-linear technique, which provided the accurate relative location of the events. Microearthquakes occurred within the shallow sediments (average depth of 3.5 km), and all appeared to be concentrated in a small volume of about 1 km3. The composite fault mechanism of the 'similar events' was computed by non-linear inversion of P-polarity readings. The maximumlikelihood solution appeared to be well constrained, and indicated a normal-faulting mechanism with the Taxis approximately oriented in a northerly direction. Interpretation of several secondary arrivals was performed by direct modelling of traveltimes and wave-motion amplitudes using a double-couple point-source model. Green's functions in a layered medium were computed using two different methods, based on ray theory (Farra 1990) and discrete wavenumber representation of the wavefield (Bouchon 1981). The study of secondary arrivals indicates a depth of 11-12 km for the top of the crystalline pre-Tertiary basement. This estimate concerns a region located at around 3-6km from the epicentres of the 'similar events', along the N290E direction. The modelling of arrival times for an S-to-P converted phase at the basement discontinuity suggests an increase of about 10 per cent in the basement seismic-wave velocity (V, = 6.9 km SKI), or, equivalently, an increase of the V, to V, ratio in the upper sediments. The increments were defined with respect to an existing reference model by AGIP (Italian Petroleum Agency). Recordings at the Minerbio station (MIN) (located about 40 km SE of the epicentral area) show converted/reflected phases at shallow crustal discontinuities and waveresonance phenomena. These path effects appear repeatedly on seismic records independently of the size of earthquakes within the magnitude range considered. Waveform and traveltime modelling of these secondary arrivals provide further constraints on the depth and velocity of main-structure discontinuities.