This part consists of 50 series of 10-11 sentences each (5 words for the sentence, on the average). The phone set balances the database. The database has associated phone, time-aligned transcriptions individual for each speaker.

The texts consist of different digit sequences. Each speaker has read 5 texts (190 words for the text, on the average) differed by digit sequence and reading style (continuos speech, isolated words). The database have been used for existing model set testing and for the refinement of models adapted for digit sequences recognition. The database has associated transcriptions obtained with the transcription-maker application. The transcriptions are common for each speaker. There is also time-aligned word transcription for the database, obtained with the our Russian voice recognition software.

Two stories (358/398 words). The phone set balances the texts. The database is used for existing model set training. The database has associated transcriptions obtained with the transcription-maker application, which were adapted by phoneticians for each speaker.

51 text set (520 words for the text, on the average). The texts based on Russian newspaper articles. The database utilization and transcribing is the same as for The Balanced Texts Speech database.

The database consists of 50 series of 10-11 sentences each (5 words for the sentence, on the average). The phone set balances the database. There are transcriptions obtained by the transcription-maker software.

4

2

2

4364

19

16

3

11057

96

68

28

16880

96

68

28

25816

35

16

19

50281

137

89

48

108398

There are 510 *.wav files for each speaker (a sentence per file). Associated with these files there are *.lab files containing time-aligned phone transcription made by Russian phoneticians.

File name mask: NNnniis.wav (NNnniis.lab),

where NN - series number,
nn -sentence number,
ii - speaker ID,
s - speaker sex.

There are 5 *.wav files containing digit sequences for each speaker. Associated with these files there are *.lab files containing phone transcription made by transcription-maker software. The transcriptions are common for each speaker. There are also *.rec files corresponding to *.wav files. These files contain time-aligned word transcription made by our Russian speech recognition software.

File name mask: 51nniis.wav (NNnniis.lab, NNnniis.rec),

where nn - text number,
ii - speaker ID.

There is *.wav file for each speaker with balanced text, he/she has read. Associated with this files there is *.lab file containing phone transcription made by transcription-maker software and adopted for the speaker by Russian phonetician.

File name mask: 53nnii[i]s.wav (nnnnii[i]s.lab),

where nn - balanced text number,
iii - speaker ID (two or three chars)
s- speaker sex

There is *.wav file for each speaker with balanced text, he/she has read. Associated with this files there is *.lab file containing phone transcription made by transcription-maker software and adopted for the speaker by Russian phonetician.

File name mask: 54nnii[i]s.wav (Nnnnii[i]s.lab),

where nn - text number
ii[i] - speaker ID (two or three chars)
s - speaker sex

There are 510 *.wav files for each speaker (a sentence per file). Associated with these files there are *.lab files containing phone transcription made by transcription-maker software and common for each speaker.

File name mask: NNnnii[i]s.wav (NNnnii[i]s.lab),

where NN - series number
nn - sentence number
ii[i] - speaker ID (two or three chars)
s - speaker sex

1

1

4

4

4

14

2040 files (510 files per speaker) are stored in a directory at Base 1: The Tables CD in individual for each speaker directories.

There are also two directories per 2040 *.lab files, containing proper time-aligned transcriptions (Latin and Cyrillic).

19 directories (individual for each speaker) per 5 files. A directory containing *.lab transcription files.

A directory containing time-aligned word transcription files (*.rec) made by transcription-maker software.

A file per speaker. 96 files in all.

A file per speaker. 96 files in all.

35 directories (individual for each speaker) per 510 files.