Giulio Maltese

An automatic technique to include grammatical and morphological information in a trigram-based statistical language model

A technique to take into account grammatical and morphological information in a trigram-based statistical language model is presented. This is automatically achieved by interpolating the trigram model (which uses sequences of words) with statistical models based on sequences of grammatical categories and/or lemmas. Such an approach reduces the effect of data sparseness in the trigram model due also to the way interpolation coefficients are chosen. With respect to trigrams, the authors obtained a significant reduction in perplexity on various texts even when combining a well-trained trigram model with a small grammatical/morphological model.<<ETX>>

Language model and acoustic model information in probabilistic speech recognition

The authors propose an approach to the estimation of the performance of the language model and the acoustic model in probabilistic speech recognition that tries to take into account the interaction between the two. It consists of a new measure, called speech decoder entropy (SDE), of joint acoustic-context information. Some results are presented for a 20000-word vocabulary recognizer. The authors discuss some limitations of the work and some suggestions for future developments.<<ETX>>

Measuring information provided by language model and acoustic model in probabilistic speech recognition: theory and experimental results

Abstract In probabilistic speech recognition it is often interesting to evaluate the contribution of the language model and that of the acoustic model. We propose an information theoretical approach which takes into account the interaction between the two sources of information. Experimental results are presented concerning the IBM prototype real-time recognizer of the Italian language based on a 20,000-word vocabulary.

Experimenting natural-language dictation with a 20000-word speech recognizer

The authors describe a newly developed real-time large-vocabulary speech recognizer for the Italian language and some preliminary experiments on its usage. Some of these experiments are aimed at evaluating voice versus keyboard as a means for entry and editing of texts. The experiments made use of a dictating-machine prototype for the Italian language, which recognizes in real time natural-language sentences built from a 20000-word vocabulary. A voice-activated editor was developed to allow the user to create, revise, file, and print documents. It is found that large-vocabulary speech recognition can offer a very competitive alternative to traditional text entry. It is likely to be well accepted even by users who have a large experience in keyboard text editing. The study has already suggested possible improvements to the man-machine interface of the current speech recognizer.<<ETX>>

Experimenting Text Creation by Natural-Language, Large-Vocabulary Speech Recognition

In the last years the probabilistic approach to speech recognition has allowed the development of high-performances large-vocabulary speech recognition systems [1] [2]. At the IBM Rome Scientific Center a speech-recognition prototype for the Italian language, based on this approach, has been built. The prototype is able to recognize in real time natural-language sentences built using a vocabulary containing up to 20000 words. [4]. Once and for all the user has to perform an acoustic training phase (about 20 minutes long), during which he is required to utter a predefined text. Words must be uttered inserting small pauses (a few centiseconds), between them. The prototype architecture is based on a personal computer equipped with special hardware. The first system we developed was aimed at a business and finance lexicon. Many laboratory tests have shown the effectiveness of the prototype as a tool to create texts by voice. After a first phase during which in-house experiments were carried on [5], the need arose to test the system in real work enviroments and for different applications. Two applications were considered: the dictation of radiological reports and of insurance company documents. Due to their characteristics, these applications seemed to be very well suited for our purposes. Since the vocabulary of the recognizer must be predefined, we had to adapt the system to the lexicon required by the new applications. The paper describes the techniques developed to efficiently adapt the basic component of the recognizer the acoustic and language models. The results obtained experimenting automatic text dictation during real work are also presented.

A 2000-word speech recognizer of Italian

A real-time speech recognition system of Italian has been developed at IBM Rome Scientific Center. It handles natural language sentences from a 20000-word dictionary, dictated with words separated by short pauses. The architecture consists of a PC/AT equipped with signal processing hardware. The paper describes the system, shows results of decoding tests and includes descriptions of the topics in speech recognition being currently investigated.