Database description
Side 1 of 1
English Language Speech Database for Speaker Recognition
(ELSDSR)
ELSDSR corpus of read speech has been designed to provide speech data for
the development and evaluation of automatic speaker recognition system.
ELSDSR corpus design was a joint effort of the faculty, Ph. D students and
Master students from department of Informatics and Mathematical Modeling
(IMM) at Technical University of Denmark (DTU). The speech language is
English, and spoken by 21 Dane, one Islander and one Canadian. Due to the
usage of this database and some realistic factors, perfect or even correct
pronunciation is not required and necessary for getting the specific and uniquely
identifiable characteristics for individual.
1. Recording Condition
The recording work has been carried out in a chamber (room 133) in building
321, 1
st
floor at DTU. The chamber is an 8.82*11.8*3.05 m
3
(width*length*height)
computer room (classroom), with 22 monitors and 34 tables. The recording is
manipulated in, approximately, the middle of this chamber, with one microphone,
one 70*120*70 cm
3
table in front of speakers. In order to deflect the reflection,
two deflection boards with measure of 93*211.5*6 cm
3
were placed at tilted
angles facing each other, and were infront of the table and speakers. For details
please see the setup drawing, drawing of the room and position of recording,
etc., in appendix.
2. Recording equipment and setup
The equipment for recording work is MARANTZ PMD670 portable solid state
recorder. PMD670 can record in a variety of compression algorithm, associated
bit rate, file format, and recording type (channels recorded) parameters. It
supports two kinds of recording format: compressed recording, which includes
MP2 and MP3; uncompressed recording, which includes linear pulse code
modulation (PCM). The recording type can be stereo, mono or digital, and the file
can be recorded into .wav .bwf .mpg or .mp3 format according to user need.
In this database, the voice messages are recorded into the most commonly
used file type--.wav. And the algorithm used is PCM. Table 1 shows the initial
setup for the recorder, for detail see PMD670 user guide.
Table 1: Recorder Setup
Setup
Input
Auto
Mark
Pre
Rec
Analog
Out
MIN
Atten
Repeat ANC
EDL
Play
Level
Cont.
S. Skip
MIC
(MONO)
OFF ON OFF 20dB OFF FLAT OFF
MANUA
L
ON
20dB
- 1
- 2
- 3
- 4
- 5
- 6
前往页