9. IPSJ SIG-SLP Corpora and Environments for Noisy Speech Recognition

9-f. Reverberant Speech Recognition Evaluation Environment (CENSREC-4)

Data DOI

https://doi.org/10.32130/src.CENSREC-4

Producer, Project

Noisy Speech Recognition Evaluation Working Group,

Special Interest Group on Spoken Language Information Processing,

Information Processing Society of Japan (IPSJ)

Contents

Common platform for evaluating independently speech recognition accuracy and speech interval detection under noisy environment.

The target evaluation framework of CENSREC-4 is distant-talking speech recognition in various reverberation environments. The data contained in CENSREC-4 are connected digit utterances as in CENSREC-1.

• Basic data set: connected digit utterances under reverberant conditions and an evaluation framework that is provided as HTK-based HMM training and recognition scripts
• Extra data set: connected digit utterances under reverberant and noisy conditions and recording data in real environments

Recording environment

In-car, Office, Meeting room, Lounge, Elevator hall, Living room, Japanese style room, Japanese style bath

(Real environment data: In-car, Office, Meeting room, Lounge)

Speaker

• Basic data set

  Training data:

  110 speakers (55 males, 55 females) 8440 utterances in total

  Test data:

  104 speakers (52 males, 52 females) 4004 utterances in total

• Extra data set

  Simulation data:

  the same as in basic data set

  Recording data in real environments:

  10 Speaker（5 males, 5 females）2536 utterances in total

Speech file format

RAW format (16kHz, 16bit, Mono, BigEndian)

Distribution media

1 DVD

Licensing

For research and development purposes only

Price

No fee

Speech sample for test listening

Basic data set: digit strings as in CENSREC-1 under reverberant conditions