Diet

A technique is described for producing smooth speech parametric representation evolutions by means of an application of traditional hidden Markov modeling techniques. It is based on the consideration that HMM training is able to locate the main acoustic events which occur during the speech process; thus the decomposition obtained can be used to reproduce the linguistic units utilized for training. This is accomplished by interpolating the state-related features values with some weighting functions, as it is done for temporal decomposition technique. Results will be given for isolated word synthesis.

A new duration intrinsic model for improved speech recognition by HMM techniques is presented. Assuming an exponentially decaying time dependency of the states loop probability, the duration density can be factorized and a path early pruning theorem demonstrated. As a consequence, computational complexity is greatly reduced with respect to explicit models, whereas recognition performances improve considerably.

The problem of finding a compact phrases list containing as many speech events as possible is addressed. The proposed algorithm performs an iterative selection from a larger set of phrases, properly phonetically transcribed. Algorithm parameters optimization has been attempted; the resulting phrases list composition is assessed by mean of histograms. Applications to speech recognition systems are envisaged.

A simple solution to the word spotting problem is experimented. Continuous Speech Recognition (CSR) techniques are adopted, with the background speech represented by a filler templates set defined through vector quantization algorithms. Quantitative results about the performed experiments are given.

ABSTRACT A syllabic strucutre model is proposed, as a tool for denning a detailed phonemic transcription. The microprosodic features (duration and loudness) of (he defined phonological-units arc estimated by statistical measures on an automatically segmented speech database. The use of these units in a text-to-specch concatenativi synthesiser is discussed.

Abstract A speech recognition system based on statistical decision theory, viewing the problem as the classical design of a decoder in a communication system framework is outlined. Statistical properties of the language are used to characterize the allowable phonetic sequence inside the words, while trying to capture allophonic phoneme features into functional-dependent acoustical models with the aim of utilizing them as word segmentation cues. Experiments prove the utility of an explicit modeling of the intrinsic ...