#### DMCA

## Generalised prior subspace analysis for polyphonic pitch transcription (2005)

Venue: | in Proc. Int. Conf. on Digital Audio Effects (DAFx |

Citations: | 10 - 3 self |

### Citations

1849 |
Independent component analysis, a new concept
- Comon
- 1994
(Show Context)
Citation Context ...ated, PSA obtained estimates of s bysmultiplying the overall spectrogram by the pseudo-inverse of thesfrequency basis functions to obtain an intial estimate of s.sIndependent Component Analysis (ICA) =-=[3]-=- then performed tosyield an improvedsestimate of s.sIn recent years, it has been proposed that sound sourcesseparation can be achieved by means of sparse decomposition in assignal dictionary [4]. The ... |

1221 | Algorithms for Non-negative Matrix Factorization
- Lee, Seung
- 2001
(Show Context)
Citation Context ...s gaussian in nature. Whilesfollowing the use of the pseudo-inverse with ICA goes some waysto solving this problem, the PSA problem is closer in formulationsto Non-negative matrix factorisation (NMF) =-=[6]-=- and Non-negativesSparse Coding (NNSC) [7].sBoth NMF and NNSC attempt to approximate a nonnegative matrix x of size n x m, such as a spectrogram, bysdecomposing it into a mixing matrix A of size n x r... |

272 | Blind source separation by sparse decomposition in a signal dictionary
- Zibulevsky, Pearlmutter
- 2001
(Show Context)
Citation Context ...(ICA) [3] then performed tosyield an improvedsestimate of s.sIn recent years, it has been proposed that sound sourcesseparation can be achieved by means of sparse decomposition in assignal dictionary =-=[4]-=-. The signal dictionary used in this researchsconsisted of a wavelet packet dictionary. More recently, it hassbeen proposed that sound source separation in single channelssignals can be carried out by... |

236 | Nonnegative matrix factorization for polyphonic music transcription
- Smaragdis, Brown
- 2003
(Show Context)
Citation Context ... basis functions chosen to represent thesoriginal data. Both methods assume that the input data is sparse insnature [7], and have been used for source separation andstranscription of polyphonic audio =-=[8,9,10]-=-. However, both suffersfrom the problem of choosing a suitable r to give the bestsinterpretation of the data, though this is less of a problem forsNMF. Further, both suffer from permutation ambiguitie... |

121 |
Separation of mixed audio sources by independent subspace analysis
- Casey, Westner
- 2000
(Show Context)
Citation Context ... each of these spectrogramsscan be represented as the outer product of an invariant frequencysbasis function, and an invariant amplitude basis function, in thesmanner of Independent Subspace Analysis =-=[2]-=-. This yields:sāā == == l j jj l j j saY 11 Ys(1)sIt is then assumed that there are known frequency basis functionssor prior subspaces apr that are good approximations to the actualssubspaces. Substit... |

57 |
Sound source separation using sparse coding with temporal continuity objective
- Virtanen
- 2003
(Show Context)
Citation Context ... basis functions chosen to represent thesoriginal data. Both methods assume that the input data is sparse insnature [7], and have been used for source separation andstranscription of polyphonic audio =-=[8,9,10]-=-. However, both suffersfrom the problem of choosing a suitable r to give the bestsinterpretation of the data, though this is less of a problem forsNMF. Further, both suffer from permutation ambiguitie... |

52 |
Polyphonic transcription by nonnegative sparse coding of power spectra
- Abdallah, Plumbley
- 2004
(Show Context)
Citation Context ...sNMF-based rule provides a better fit to the underlying data. Also,sin most cases 50 iterations was found to give sufficientsconvergence.sAnother useful update rule for s is that proposed bysAbdallah =-=[11]-=-. It should be noted that the use of that update rulesrequires the use of a power spectrogram as opposed to asmagnitude spectrogram, and that the prior subspaces have to besmodified in accordance with... |

46 | Separation of sound sources by convolutive sparse coding
- Virtanen
- 2004
(Show Context)
Citation Context ...ception of a small number of low amplitude notes. This was after thresholding and elimination of shortsduration activations of s. It is possible that some form of perceptual weighting as described in =-=[16]-=- on the input spectrogram maysbe of use in recovering these notes. The success of this test showssthat the algorithm can function on real-world signals, and that assynthetic prior can be used to attem... |

35 | Drum transcription with nonnegative spectrogram factorization.
- Paulus, Virtanen
- 2005
(Show Context)
Citation Context ...n that the new algorithm has outperformed the original PSAsalgorithm.sIt should be noted that a similar reformulation of thesPSA algorithm described above was arrived at independently bysPaulus et al =-=[12]-=-. However, for the purposes of pitched instrumentstranscription, both reformulations suffer from the need forsansindividual prior subspace for each note present. Methods tosovercome this problem are p... |

33 |
Non-negative sparse coding,ā in Neural Networks for Signal Processing
- Hoyer
(Show Context)
Citation Context ...use of the pseudo-inverse with ICA goes some waysto solving this problem, the PSA problem is closer in formulationsto Non-negative matrix factorisation (NMF) [6] and Non-negativesSparse Coding (NNSC) =-=[7]-=-.sBoth NMF and NNSC attempt to approximate a nonnegative matrix x of size n x m, such as a spectrogram, bysdecomposing it into a mixing matrix A of size n x r, and a set ofsfeature vectors s, of size ... |

30 |
Automatic drum transcription and source separation.
- FitzGerald
- 2004
(Show Context)
Citation Context ...orithmsfor the purposes of polyphonic pitch transcription.s1. INTRODUCTIONsPrior Subspace Analysis (PSA) was first proposed as a techniquesfor transcription and sound source separation of drum sounds =-=[1]-=-,sand was found to be successful at tackling the transcription ofscertain types of drum sounds. However, the method was not suitable for the transcription of pitched instruments, as it required ansind... |

22 |
Music transcription with ISA and HMM
- Vincent, Rodet
- 2004
(Show Context)
Citation Context ...notes from the same instrumentsoccur over the course of a spectrogram. Previous work attemptingsto deal with this includes the non-linear Independent SubspacesAnalysis model proposed by Vincent et al =-=[13]-=-.sIn this model,schord spectra are represented as sums of note power spectra, andsnote spectra are represented as sums of instrument dependant logpower spectra. Note durations are then modeled using H... |

15 |
Shifted non-negative matrix factorisation for sound source separation
- FitzGerald, Coyle
(Show Context)
Citation Context ... contains thesamplitude basis function, of size 1 x m associated with translationsmatrix Tj of size n x n. An algorithm which attempts to learn bothsA and sj from an input spectrogram is described in =-=[14]-=-.sThe utility of this signal model can be seen in that a single basis function can now be used to model a pitched instrumentsand can be seen as a means of generalising the PSA model to dealswith pitch... |

15 | Mir in matlab: The midi toolbox. In:
- Eerola, Toiviainen
- 2004
(Show Context)
Citation Context ...vious example, and the samesprior, a piano-note of pitch G5 was used. A pianoroll plot of thesmidifile used to generate the audio signal is shown in Figure 3.sThis was generated using the Miditoolbox =-=[15]-=-. Figure 4 thensshows the output of the generalised PSA algorithm. Rememberingsthat a translation of 0 corresponds to note G5, it can be seen thatsProc. of the 8th Int. Conference on Digital Audio Eff... |

13 | Monaural source separation using spectral cues
- Pearlmutter, Zador
- 2004
(Show Context)
Citation Context ...tion in single channelssignals can be carried out by fitting an overcomplete signal dictionary to the signal, in conjunction with knowledge of spectralscues such as the head related transfer function =-=[5]-=-.sThe above research suggests a different view of thesprior subspaces used in PSA, namely that the prior subspaces are assignal dictionary, albeit a very undercomplete signal dictionary.sThe PSA probl... |

2 | Nonnegative matrix factorization: a possible way to learn sound dictionaries
- Asari
- 2005
(Show Context)
Citation Context ... basis functions chosen to represent thesoriginal data. Both methods assume that the input data is sparse insnature [7], and have been used for source separation andstranscription of polyphonic audio =-=[8,9,10]-=-. However, both suffersfrom the problem of choosing a suitable r to give the bestsinterpretation of the data, though this is less of a problem forsNMF. Further, both suffer from permutation ambiguitie... |