Abstract | ||
---|---|---|
In this work, we propose an objective measure to assess the amount of musical noise that results from speech enhancement algorithms. The algorithms can result in nonsmooth suppression of background noise which in turn translates to isolated regions of high energy, referred to as musical noise. We propose to identify such regions by combining time-frequency (TF) bins associated through connectivity along with additional properties of these regions such as area, aspect ratio and total energy. The objective measure proposed is based on density of such regions. The effectiveness of the proposed measure is studied by correlating it with subjective assessment of listeners using enhanced speech of various algorithms. |
Year | DOI | Venue |
---|---|---|
2019 | 10.1109/NCC.2019.8732246 | 2019 National Conference on Communications (NCC) |
Keywords | Field | DocType |
Musical Noise,Speech Enhancement,connected TF regions | Speech enhancement,Background noise,Musical,Computer science,Speech recognition,Time–frequency analysis,High energy | Conference |
ISBN | Citations | PageRank |
978-1-5386-9286-8 | 0 | 0.34 |
References | Authors | |
0 | 4 |
Name | Order | Citations | PageRank |
---|---|---|---|
Ajey Saligrama | 1 | 0 | 0.68 |
H. G. Ranjani | 2 | 11 | 3.05 |
Rangarao Muralishankar | 3 | 26 | 8.13 |
H. N. Shankar | 4 | 10 | 4.93 |