In brackets there are the names of people who you should contact to get detailed information about the software.
Resources
JURISDIC Database
(Grażyna Demenko)
The JURISDIC database is intended to provide material for both training and testing of speech dictation of common and legal texts, including isolated word systems, word-spotting systems and vocabulary independent systems which use either whole word or subword modeling approaches.
The JURISDIC common specification is a mixture of semi-spontaneous (controlled dictation) and read/dictated speech.
The JURISDIC specification for Polish is based on the general language features and peculiarities of Polish on the different linguistics as well as phonetics levels. This results in recording session duration of approx. 60 minutes for one speaker. Number of speakers: 1000.
Lexicon for application in speech technology
(Agnieszka Wagner)
The lexicon for application in speech technology systems and especially in a speech dictation and recognition system Jurisdic has been created according to the LC-STAR project specifications. It consists of three parts: common words, proper names and special application words. Additionally, a lexicon based on frequency dictionaries has been created in order to ensure high lexical coverage: the procedures used to built it are described in detail in the Report.
Software
Word spotting
(Stefan Grocholewski)
There is a huge number of recordings in police archives which is only an acoustic signal. Finding the necessary information via listening is very time-consuming, that is why the need for automatic browsing of audio recordings has risen. At the moment (fall 2007) first experiments are conducted with the use of the HMM based system. The vocabulary in such a system consists of all phonemes and the key words. We elaborated the method to eliminate the improper candidates. In Figure 1 only two candidates are chosen from about 30.
Figure 1: The working window of the WordSpotter program.
Automatic speaker identification and verification
(Stefan Grocholewski)
The aim of our project is to create a necessary tool which would make full or at least partial automatic speakers recognition possible. At the moment (fall 2007) basic GMM based modules are being created. Basic problems are: elimination of the influence of transmission channels and acoustic background, and the most important, elimination of the influence of emotions on phonetic/acoustic parameters of the recording. In Figure 2 a working window of the Speaker Identification program is showed.
Figure 2: The working window of the Speaker Identification program.
Polphone (Marcin Szymański or Marek Lange)
Pitch Line (Jerzy Ogórkiewicz)
Annotation System for the Polish ASR database for Polish Platform for Homeland Security (PPHS, pl. PPBW)
(Katarzyna Klessa)
"PPBW Annotation Database Manager" was designed based on the
Client-Server architecture using MSDE 2000, and Windows 2003 Server. The
software is in charge of the sound and label files, text files, speaker
information, backup copies, work time statistics as well as multi-user
management and lexicon search. For the purpose of segmenting and
labeling speech Transcriber, an open-source tool, was integrated in the
system.
The program for checking and assesing the database of recordings (Daniel Śledziński)
Boss (Marcin Szymański)
Automatic Close Copy Speech (ACCS) synthesis
(Jolanta Bachan)
Automatic Close Copy Speech (ACCS) synthesis system based on MBROLA diphone synthesis . The design of the Automatic Close Copy Speech synthesiser allows the system to be used with languages from all over the world. The requirements for the system to work with a given language are:
- an MBROLA voice for a given language exists;
- a corpus of recordings and annotations of these recordings on the phoneme level in the TextGrid format is available;
- a mapping of the annotation phoneme labels inventory into MBROLA voice phoneme labels is developed.
Automatic Close Copy Speech (ACCS) synthesis demo:
Speech Perception Tests for Children with a Cochlear Implant
(Jolanta Bachan)
A battery of speech perception tests for children with a cochlear implant. Tests examine children's perceptive and linguistic skills making use of acoustic signals only. The tests are designed for children who are able to comprehend speech, but who may be unable to give verbal responses.
Recording equipment and software - Watch the movie! (Jerzy Ogórkiewicz)
