Project description: High Noise Speech Corpus

---

◦ 100 speakers were recorded through one common microphone to pick up air 
conduction sound and three types of vibration microphones to pick up bone 
conduction sound - headgear type, ear type, neck type - simultaneously in 2 
opposite environments (in the states of noise propagation and no noise 
propagation in the anechoic sound field): no noise and high noise to be used 
for research into improvement of speech recognition. 40 speakers were recorded 
in no-noise environment (noise sound pressure: 0dBA), 20 speakers in high noise 
environment of played 60 dBA white noise and 20 speakers in high noise 
environment of played 80 dBA (SPL) white noise. 10 speakers who participated in 
recordings in no-noise environment also participated in recordings in 60 dBA 
and 80 dBA random noise environments to be used for the analysis of Lombard 
effect depending on noise levels
◦ A total of 100 speakers (Male: 50, Female: 50)
◦ Soundproof room
 - background noise level: 20dB(A) 
 - reverberation time: 0.4sec(@500Hz)
◦ Microphone
 - primary microphone: Shure 565SD (Typical Dynamic Type)
 - secondary microphones:
 Temco HG-17 (Bone Conduction Headgear Type)
 Temco EM-7 (Bone Conduction Ear Microphone)
 Fostec FSN-101A (Neck Vibration Pick-up Type)
◦ Sampling and data format
 - 16,000 Hz, 16 bit Windows wave format
◦ Prompts
 - 452 phonetically balanced words
 - 5 sentences containing all the Korean phonemes (for all the speakers)
◦ Prompts Sheet
 - 2 sets of words
 - 5 sentences
◦ Per speaker
 - 1 set of 226 words 
 - 5 sentences