Showing papers in "Computer Speech & Language in 1996"

TL;DR: An adaptive statistical language model is described, which successfully integrates long distance linguistic information with other knowledge sources, and shows the feasibility of incorporating many diverse knowledge sources in a single, unified statistical framework.

TL;DR: This paper examines the maximum likelihood linear regression (MLLR) adaptation technique, which has been applied to the mean parameters in mixture-Gaussian HMM systems and is extended to also update the Gaussian variances and re-estimation formulae are derived for these variance transforms.

TL;DR: This paper used decision trees to predict prosodic labels for text-to-speech systems, such as accent location, symbolic tones, and relative prominence level, from text tagged with part-of-speech labels and marked for prosodic constituent structure.

TL;DR: This work presents the VariableN-gram Stochastic Automaton (VNSA) language model that provides a unified formalism for building a wide class of language models and shows that the VNSAs are well suited for those applications where speech and language decoding cascades are implemented through weighted rational transductions.

TL;DR: Although the proposed transform has been derived heuristically—namely, to be optimal in the perceptual frequency scale in Gabor-sense and to perform a 1 CB speech analysis—it appears that this is a self-invertible, overcomplete, shiftable transform.

TL;DR: Results show effective unsupervised speaker adaptation using only 5 s calibration speech and the method reduced the error rate by 8·5% compared with the conventional speaker-independent speech recognition method.

TL;DR: Current results for letter-to-phoneme conversion are at least as good as the best reported so far for a data-driven approach, while being comparable in performance to knowledge-based approaches.

TL;DR: In this article, a speaker recognition method based on hidden Markov model (HMM) composition was proposed, which combines a speaker HMM and a noise source HMM into a noise-addressed speaker HMM with a particular signal-to-noise ratio (SNR).

TL;DR: A numerical simulation of laryngeal flow was developed to study flow patterns and pressure and velocity waveforms in a model of the oscillating glottis, indicating that with this simulation of the entire flow field, periodic velocity and pressure fields exist throughout the larynx.

TL;DR: The improved LID system was evaluated on an 11-language task, and performance reached 13·3% and 26·2% for utterances averaging 45 s duration and 10 s duration, respectively, which shows the importance of the issues addressed in this paper for language identification.

TL;DR: Comparisons on a speaker-independent E-set database show that the new model, without optimization on the dependence structure, achieves better performance than the standard HMM, the bigram HMM and the linear-predictive H MM, all in comparable or smaller parameter sizes.

TL;DR: It was found that adaptation using this initial model reduces the phoneme recognition error rate from 22·0% to 17·7%, showing the effectiveness of using speaker similarity information as a priori information.

TL;DR: A new search algorithm for speech recognition which applies the monotone graph search procedure to the problem of building a word graph is described and it is shown how the search can be made to run very quickly if the 1-phone lookahead assumption holds.

TL;DR: By using the formulation of the missing-data problem, a general framework for statistical acoustic modelling of speech is presented and a bi-directional hidden Markov modelling approach for speech recognition is studied.

TL;DR: This work explicitly model the time evolution of an observed trajectory by the sum of a first order AR process and a mean component and reduces the recognition error rate on an 850-word vocabulary continuous speech recognition task.