9. IPSJ SIG-SLP Speech Recognition Evaluation Environment

9-a. Noisy Speech Recognition Evaluation Environment (CENSREC-1 ⟨AURORA-2J⟩)

Data DOI

https://doi.org/10.32130/src.CENSREC-1

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. "CENSREC" stands for "Corpus and Environment for Noisy Speech RECognition."

CENSREC-1 (AURORA-2J) is a Japanized version of the AURORA-2 database.

• Training data (clean training dataset and multi-condition training dataset)
• Test data (Testset A, B and C)

Speaker

Training data:

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

Test data:

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

Speech file format

RAW format (8kHz, 16bit, Mono, BigEndian)

Distribution media

1 DVD

Licensing

For research and development purposes only

Price

No fee

Speech sample for test listening

Digit strings same as the AURORA-2 (but it was uttered in Japanese)

Noise environment: Subway

Utterance: 1 (/ichi/)

Clean  / SNR 20dB  / SNR 15dB  / SNR 10dB  / SNR 5dB  / SNR 0dB  / SNR -5dB