Out-Of-Breath Speech Database Analysis

In this work, a new stressed speech database is recorded.
This recorded database is named as out-of-breath speech
(OBS) database. The database contains three classes of speech corresponding to three different levels of breath emission. These three classes are out-of-breath speech, low out-of-breath speech and normal speech. The out- of-breath speech is defined as the speech produced with excessive emission of breath, where as low out-of-breath speech contains lower level of breath emission compared to the out-of-breath speech but higher than the normal speech. Ten male speakers participated in the recording of the OBS database. The speakers are research schol- ars of Indian Institute of Technology Guwahati (India), and are knowledgeable and

The subjective evaluation helps determine human ability for classification of out-of-breath speech from the normal speech. For listening purpose, we have used same headset
(HP Headphone with Microphone-B4B09PA), which was used for recording purpose. The headset contains both the microphone for recording and the headphone for listening.
Before listening test, we have provided training example to the listeners so that the listeners get acquainted about the three classes of OBS database. Two different tests
[29] are performed for assessment of out-of-breath speech
(OBS) database. In the first test, 200 wave-files are cho- sen. Out of 200 wave-files, 100 files are the normal speech and the remaining are the out-of-breath speech. These
200 wave-files are ordered in random manner which is not known by the listeners. The listeners are requested to identify if the wave-file has normal speech or out-of-breath speech. The average classification rate obtained with this subjective listening test is approximately 89%. In the sec- ond test, 50 wave-files are chosen from each of the three classes (normal speech, out-of-breath speech and low out- of-breath speech). The listeners evaluate a series of set