Grazyna Demenko

Polish unit selection speech synthesis with BOSS: extensions and speech corpora

This article presents research and development aimed at creating a Polish speech database for speech synthesis and adapting BOSS (The Bonn Open Synthesis System) to the Polish language. First of all, the linguistic background for the design of Polish spoken resources for unit selection is presented, together with the presentation of the applied transcription and annotation methods. The next section details the assumptions and the structure of the Polish corpus and its segmental and prosodic annotation. Then, the linguistic features used in duration modelling and the selection of adequate speech units of two Polish modules in BOSS are reported: the duration prediction module (the description is accompanied by a concise overview of Polish duration modelling for speech technology purposes) and the cost functions module. Finally, the results of two kinds of perception tests are discussed: the first is a preference test aimed at the evaluation of synthesized speech obtained using three variants of speech signal segmentation (automatic, semi-automatic and manual) and the second is a mean opinion score test carried out to provide a preliminary assessment of the synthesized speech quality attained with the Polish version of the BOSS synthesizer. The closing chapter summarizes future perspectives and challenges for the Polish TTS (text-to-speech) and further developments of BOSS for Polish.

Polish Speech Dictation System as an Application of Voice Interfaces

This paper presents the results of the project realized at PSNC and supported by The Polish Ministry of Science and Higher Education – “Integrated system of automatic speech-to-text conversion based on linguistic modeling designed in the environment of the analysis and legal documentation workflow for the needs of homeland security”, aiming at developing a Polish speech dictation (or Large Vocabulary Continuous Speech Recognition, LVCSR) system designed with the use of a phonetically controlled large vocabulary speech corpus and a large text corpora. The functions of the resulting system are outlined, the software architecture is presented briefly, then the example applications are demonstrated and the recognition results are discussed.

Voice as a Key

Biometric methods numerous advantages are known very well for every computer security specialist and are in general highly valued. One of the most interesting methods used for biometric protection of computer system (and other objects) is speech parameters extraction, analysis and recognition. Problem under consideration is concentrated on methods showing, how to distinguish between voices of particular people and how to do it independent on the text pronounced by recognized person. Methods used for this purpose are complementary ones in comparison with widely used speech recognition goals, where we must understand the merit content of the utterance disregarding individual features of the voices of different speakers taken into account. The paper shows general problems related to biometric speaker identification and gives general outline, how to use voice as a key for guarded systems.

Speech Man-Machine Communication

For modern man-machine communication traditional methods based on a keyboard and a mouse are definitely insufficient. Especially thinking about poor and bad-qualified citizens of Information Society we must try to find the easiest and most comfortable method for man-machine communication. This optimal method of communication going from a man to a machine is (or should be) speech communication. In the paper some general remarks about speech man-machine communication are presented and some problems connected with this area of technological activity discussed. On the basis of such discussion the survey of speech recognition systems is shown and compact analysis of the most important and most up-to-date scientific and technological problems related to such systems is presented. Final conclusions are directed both toward presentation of existing speech recognitions systems as well as toward forecasting of solutions, which will be available in the near future.

Usefulness of low redundancy speech tests with dyslexic children

Abstract Objectives: Developmental dyslexia in children can affect auditory and linguistic skills. Due to the impairment of hearing discrimination, attention, memory and perception, dyslexia causes the inability to process and interpret linguistic and verbal information effectively. Standard audiological examination is much less useful than low redundancy speech tests and electro-physiological examination, which are the only objective measures of central auditory processing. The aim of the study was to assess the usefulness of low redundancy speech tests and dichotic tests in the diagnosis of developmental dyslexia in children. Methods: Ten children aged 7–15 years were tested by a dichotic numeral test, a dichotic verbal minimal-pair test, and a Calearo test. Results: The experiments demonstrated that the dyslectic group achieved worse results in the above tests compared to the control group. This suggests information exchange disorders between the hemispheres, a lack of synchronization of acoustic perception in dichotic tests, and damage of the structures responsible for central auditory processing in the dominating hemisphere. Conclusions: The results of our study confirm the existence of information exchange disorders between the hemispheres and the lack of synchronization of acoustic perception in dichotic tests, where the signal is presented to both ears simultaneously. Much worse results in low redundancy tests reveal, moreover, damage of the structures responsible for central auditory processing in the dominating hemisphere. Altogether, this research has proved the usefulness of the conducted tests in the diagnostics of central auditory processing disorders in patients with dyslexia.

Dalsze badania nad testami mowy utrudnionej

INTRODUCTION The purpose of speech audiometry is to extend the diagnostics of central auditory processes. First audiometric tests were monaural tests introduced in the fifties by Bocca and Calearo to diagnose Central Auditory Processing Disorders (CAPD). Monaural tests are characterized by low redundancy, which allows for deciphering a given word or phrase only if the amount of accessible information exceeds a required minimum. Until now, a new monaural filtered speech test has been drawn up at the Department of Phoniatrics and Audiology. The purpose of the research was to draw up and assess the effectiveness of monaural low redundancy tests, in which the acoustic signal has been compressed or noise has been added. MATERIALS AND METHOD 35 people with normal hearing, aged 16-60, have been examined. The test material included: a compressed speech test, in which the speech signal has been accelerated by 25% with the use of Cool Edit Pro 2.0; the pauses between the words have not been shortened a speech-in-noise test, in which the noise has been added with the use of Cool Edit Pro 2.0. The signal/noise ratio has been set at the level of 10 dB. The examination has been conducted at three different volume levels: 25, 35 and 45 dB. CONCLUSIONS Best results have been achieved at the volume level of 45 dB (above hearing threshold); There were differences in results at volume levels 25 dB and 35 dB in compressed speech test and at volume level 25 dB in speech in noise. Both tests have shown the influence of age on the test result which suggests undiagnosed CAPD among people (with normal hearing) aged over 55.Summary Introduction The purpose of speech audiometry is to extend the diagnostics of central auditory processes. First audiometric tests were monaural tests introduced in the fifties by Bocca and Calearo to diagnose Central Auditory Processing Disorders (CAPD). Monaural tests are characterized by low redundancy, which allows for deciphering a given word or phrase only if the amount of accessible information exceeds a required minimum. Until now, a new monaural filtered speech test has been drawn up at the Department of Phoniatrics and Audiology. The purpose of the research was to draw up and assess the effectiveness of monaural low redundancy tests, in which the acoustic signal has been compressed or noise has been added. Materials and method 35 people with normal hearing, aged 16–60, have been examined. The test material included: a compressed speech test, in which the speech signal has been accelerated by 25% with the use of Cool Edit Pro 2.0; the pauses between the words have not been shortened a speech-in-noise test, in which the noise has been added with the use of Cool Edit Pro 2.0. The signal/noise ratio has been set at the level of 10 dB. The examination has been conducted at three different volume levels: 25, 35 and 45 dB. Conclusions Best results have been achieved at the volume level of 45 dB (above hearing threshold); There were differences in results at volume levels 25 dB and 35 dB in compressed speech test and at volume level 25 dB in speech in noise. Both tests have shown the influence of age on the test result which suggests undiagnosed CAPD among people (with normal hearing) aged over 55.