CITATION — REFERENCE ENTRY
Speech segregation based on sound localization — The Journal of the Acoustical Society of America
- Key
- roman2003-localization
- Authors
- Roman, Nicoleta; Wang, DeLiang; Brown, Guy J.
- Issued
- 2003-10
- Type
- article-journal
- Container
- The Journal of the Acoustical Society of America
- Volume
- 114
- Issue
- 4
- Pages
- 2236-2252
Raw CSL JSON
{
"DOI": "10.1121/1.1610463",
"URL": "https://pnlwang.github.io/papers/RWB.jasa03.pdf",
"PMID": "14587621",
"page": "2236-2252",
"type": "article-journal",
"issue": "4",
"title": "Speech segregation based on sound localization",
"author": [
{
"given": "Nicoleta",
"family": "Roman"
},
{
"given": "DeLiang",
"family": "Wang"
},
{
"given": "Guy J.",
"family": "Brown"
}
],
"issued": {
"date-parts": [
[
2003,
10
]
]
},
"volume": "114",
"container-title": "The Journal of the Acoustical Society of America"
}
Claims
-
The system separates target speech from interference with a supervised learning approach using interaural time and intensity differences.
"This paper describes a novel, supervised learning approach to speech segregation, in which a target speech signal is separated from interfering sounds using spatial localization cues: interaural time differences (ITD) and interaural intensity differences (IID)."
-
An ideal binary mask element is 1 where target energy exceeds interference energy in a time-frequency unit and 0 otherwise; such a mask gives a performance ceiling for all binary masks.
"An element in the mask is 1 if the corresponding T–F unit contains target energy that is stronger than interference energy, and 0 otherwise. ... In addition, from a theoretical ASA perspective, an ideal binary mask gives a performance ceiling for all binary masks."
-
The algorithm improved intelligibility scores in the conditions tested, with the improvement growing as SNR decreased, reaching 61% at -10 dB.
"The general finding is that our algorithm improves the intelligibility score for the tested conditions. The improvement becomes larger as the SNR decreases (61% at −10 dB), even though the algorithm introduces more target distortions at lower SNR levels."
-
Evaluated by signal-to-noise ratio and automatic speech recognition, the system produced masks very close to ideal binary masks.
"A systematic evaluation in terms of signal-to-noise ratio as well as automatic speech recognition performance shows that the resulting system produces masks very close to ideal binary ones."
Available in