#### DMCA

## Extension of Sparse, Adaptive Signal Decompositions to Semi-Blind Audio Source Separation

### Cached

### Download Links

Citations: | 4 - 3 self |

### Citations

3136 |
A Wavelet Tour of Signal Processing
- Mallat
- 1998
(Show Context)
Citation Context ...eld sparser representations and superior performance compared to non-adaptive transforms with constant block lengths [9, 10]. Such methods include adaptive, dyadic lapped orthogonal transforms (LOTs) =-=[6]-=- and adaptive, non-dyadic LOTs, which give better performance in return for higher computational complexity [7]. We recently introduced MPEG-like LOTs, which aim for a trade-off between improving comp... |

164 | Underdetermined blind source separation using sparse representations
- Bofill, Zibulevsky
- 2001
(Show Context)
Citation Context ...nsform of s(n), where 0 ≤ m < N. We assume that exactly two sources are active at each m because this gives better performance than the simpler binary masking case which allows only one active source =-=[1, 10]-=-. The set of both source indices contributing to X(m) is denoted by Jm = {j : Sj(m) ̸= 0}, and is called the local activity pattern at m. Given a particular Jm, Equation (1) then reduces to a determin... |

42 | Oracle estimators for the benchmarking of source separation algorithms, Signal Processing 87 (8
- Vincent, Gribonval, et al.
- 2007
(Show Context)
Citation Context ...n, then the problem is called semi-blind. If both A and s(n) are known, then we can determine upper performance bounds; this ideal oracle estimation case is useful for algorithm benchmarking purposes =-=[10]-=-. Underdetermined audio source separation is typically addressed by timefrequency (TF) masking, which assumes that we can transform x(n) by a linear, invertible TF transform so that the sources overla... |

16 |
Piecewise linear source separation
- Gribonval
(Show Context)
Citation Context ...source separation is typically addressed by timefrequency (TF) masking, which assumes that we can transform x(n) by a linear, invertible TF transform so that the sources overlap as little as possible =-=[4]-=-. State-of-the-art methods have the potential to yield sparser representations and superior performance compared to non-adaptive transforms with constant block lengths [9, 10]. Such methods include ad... |

7 | Blind criterion and oracle bound for instantaneous audio source separation using adaptive timefrequency representations
- Vincent, Gribonval
- 2007
(Show Context)
Citation Context ...verlap as little as possible [4]. State-of-the-art methods have the potential to yield sparser representations and superior performance compared to non-adaptive transforms with constant block lengths =-=[9, 10]-=-. Such methods include adaptive, dyadic lapped orthogonal transforms (LOTs) [6] and adaptive, non-dyadic LOTs, which give better performance in return for higher computational complexity [7]. We recen... |

6 | Best basis search in lapped dictionaries
- Huang, Pollak, et al.
- 2006
(Show Context)
Citation Context ...brary and is given by L = ⋃ λ∈Λ Bλ . 4 Fast and Flexible Partitioning Schemes For any additive function C, we can use dynamic programming to determine the BOB which minimises C( ̂ S) over all B λ ∈ L =-=[3, 6]-=-. Such algorithms jointly estimate the local activity patterns Jm according to (5) and find the best orthogonal basis which minimises the ℓ 1 norm given by (4) according to ̂λ = arg min C( λ∈Λ ̂ S) . ... |

5 |
Identifying Single Source Data for Mixing Matrix Estimation
- Bofill
(Show Context)
Citation Context ...s contributing to X(m) is denoted by Jm = {j : Sj(m) ̸= 0}, and is called the local activity pattern at m. Given a particular Jm, Equation (1) then reduces to a determined system: X(m) = AJmSJm (m) , =-=(2)-=- where AJm is the 2×2 submatrix of A formed by taking columns Aj, and SJm(m) is the subvector of S(m) formed by taking elements Sj(m), whenever j ∈ Jm. Once Jm has been estimated for each m we estimat... |

4 |
Oracle evaluation of flexible adaptive transforms for underdetermined audio source separation
- Nesbit, Plumbley, et al.
- 2008
(Show Context)
Citation Context ...ngths [9, 10]. Such methods include adaptive, dyadic lapped orthogonal transforms (LOTs) [6] and adaptive, non-dyadic LOTs, which give better performance in return for higher computational complexity =-=[7]-=-. We recently introduced MPEG-like LOTs, which aim for a trade-off between improving computation time, and decreasing artefacts at window boundaries and improving performance, and evaluated them in or... |

4 | Benchmarking flexible adaptive time-frequency transforms for underdetermined audio source separation
- Nesbit, Vincent, et al.
- 2009
(Show Context)
Citation Context ...introduced MPEG-like LOTs, which aim for a trade-off between improving computation time, and decreasing artefacts at window boundaries and improving performance, and evaluated them in oracle contexts =-=[8]-=-. In this paper, we extend this previous work by evaluating them in semi-blind contexts. 2 Time-Frequency Masking Let us denote by X(m) = (X1(m), X2(m)) the TF transform of x(n), and let S(m) = (S1(m)... |