SZYMON DRGAS- zobacz i pobierz wszystkie publikacje autora SZYMON DRGAS w naszych czasopismach

Wyniki 1-5 spośród 5 dla zapytania: authorDesc:"SZYMON DRGAS"

» Speech recognition by means of feature extraction method based on slope transformation assisted

Current automatic speech recognition (ASR) systems achieve already acceptable recognition accuracy for clean speech. However their performance degrades, if they are subjected to speech corrupted with noise present in practical environments. It has been observed that additive white noise severely degrades the performance of mel-cepstra based recognition systems [8]. This problem becomes espec[...] więcej»
w zeszycie ELEKTRONIKA - KONSTRUKCJE, TECHNOLOGIE, ZASTOSOWANIA 2007/4

KUP TERAZ » KUP TERAZ (SMS) »

» Speaker identification based on Gaussian mixture model - experiments with Polish language utterances

ADAM DĄBROWSKI

SZYMON DRGAS

DAMIAN CETNAROWICZ

IRENA CHMIELEWSKA

The speech signal conveys several levels of information, however most important is the message coded with the words. Furthermore speech signal conveys information about speaker identity, emotions, health etc. Automatic speaker recognition is a technology that is capable to identify a person who spoke (speaker identification) or to authenticate the speaker's claimed authenticity (speake[...] więcej»
w zeszycie ELEKTRONIKA - KONSTRUKCJE, TECHNOLOGIE, ZASTOSOWANIA 2008/4

KUP TERAZ » KUP TERAZ (SMS) »

» Koncepcja multimedialnej bazy danych do rejestracji i przeszukiwania rozmów na telefony alarmowe

JULIAN BALCEREK

SZYMON DRGAS

ADAM DĄBROWSKI

ADAM KONIECZKA

Rozmowy na telefony alarmowe są w Polsce rejestrowane, jednak ze względu na ich dużą liczbę, w chwili obecnej nie jest jeszcze możliwe odnalezienie odpowiednich informacji (np. potrzebnych do identyfi kacji mówcy) bez wykorzystania znacznych zasobów ludzkich, a więc poświęcenia temu zadaniu dużych nakładów czasowych i fi nansowych. Z tego powodu jednym z celów prowadzonego przez autorów projektu naukowego jest opracowanie bazy danych zawierającej informacje o rozmowach i narzędzia do wielopoziomowego wspomagania rozpoznawania mówcy, tj. metody wykorzystujące zarówno cechy głosu, jak i lingwistyczne cechy wypowiedzi mówcy. Narzędzia te są oparte na implementacji algorytmów wielopoziomowej analizy sygnałów i anotacji danych wykonywanej przez operatora telefonu i/lub automatyczni[...] więcej»
w zeszycie ELEKTRONIKA - KONSTRUKCJE, TECHNOLOGIE, ZASTOSOWANIA 2010/3

KUP TERAZ » KUP TERAZ (SMS) »

» Fast speaker recognition based on short Polish sequences

TOMASZ MARCINIAK

Among key issues determining effectiveness and speed of biometric identification systems there is a proper selection of acquisition parameters and representation of the input data. Typically biometric systems are based on the analysis and recognition of various images such as fingerprint, face, iris, ear, hand geometry, etc. [1, 2]. However, nowadays speaker recognition techniques are becoming more and more important. Identification techniques based on acoustic signals (voice) is yet holding about a 3% share in the commercial biometric market only [3]. However, it should be noted that the speaker identification has a number of advantages and can be used to authorization access for many services and systems such as voice dialing options, telephone banking, shopping by phone, database access, voicemail, information services, access to restricted zones, access to computers, etc. In contrast to systems based on image recognition, speaker recognition easier detects sex and nationality. It may also be a part of complex multimodal biometric systems examining many biometric features thus guarantying more effective identification. Speaker recognition techniques can be divided into two types: verification and identification [4, 5]. Speaker verification consists in acceptance or rejection of the speaker. Speech after parameterization is compared with the reference model. Depending on certain diagnosis threshold, the speaker is accepted or rejected. Verification is simpler than the second task namely identification, which consists in recognition which person from a set of the registered people speaks. Parameters of the input speech signal are compared with the base parameters of the reference N-models. Then the maximum selector shows the greatest similarity to the reference model and selects the appropriate speaker ID. Speaker recognition methods can be divided into the following main categories [6]: - text-dependent - speaker recogni[...] więcej»
w zeszycie ELEKTRONIKA - KONSTRUKCJE, TECHNOLOGIE, ZASTOSOWANIA 2011/5

KUP TERAZ » KUP TERAZ (SMS) »

» Database and recording system for registration, maintaining and fast searching of emergency telephone calls

JULIAN BALCEREK

Maintenance of telephone calls is a very important task for emergency services like for the Police, Ambulance Service, and Fire Brigades [1]. An important task consists, among others, in a search for calls of same person over a relatively long period of time (of months or years) and large area. Huge number of the recorded calls to be analyzed in such a case hampers or even makes realization of this task impossible not only in real-time (i.e., during the call, which typically lasts about 1.2 minutes only) but also off-line. Furthermore, it is also quite difficult to automatically decide if a present call is related to the already reported case. The described functionalities are very helpful to prevent unnecessary interventions under fake submissions or multi reporting of the same case, which unfortunately occur quite often. In order to offer the described functionalities, we are creating an additional database in parallel to the standard one (that of pure, whole call recordings), which contains concise but unambiguous information about the calls and speakers, and we are preparing tools for efficient and fast data searching. These tools are based on implementation of multilevel signal analysis and on metadata, which are automatically or manually annotated (i.e., by the telephone operator) for the call in real-time and/or during the call post-processing. Some data should be won manually as e.g. annotations of speaker characteristics based on the call transcriptions. Proposed tools enable efficient data search among the calls. It is possible to recall the most similar voices from the database using multilevel speaker analysis starting with the voice features and ending with the linguistic features. For more effective searching we use metadata in a complex but coherent system based on semantic data analysis joint with the numeric data analysis, and correlations among various features. General, initial concept of the database sys[...] więcej»
w zeszycie ELEKTRONIKA - KONSTRUKCJE, TECHNOLOGIE, ZASTOSOWANIA 2011/5

KUP TERAZ » KUP TERAZ (SMS) »

Strona 1

r e k l a m a