## Bayesian Analysis of Polyphonic Western Tonal Music (2006)

Venue: | JOURNAL OF THE ACOUSTICAL SOCIETY OF AMERICA |

Citations: | 17 - 3 self |

### BibTeX

@ARTICLE{Davy06bayesiananalysis,

author = {Manuel Davy and Simon J. Godsill and Jérôme Idier},

title = {Bayesian Analysis of Polyphonic Western Tonal Music},

journal = {JOURNAL OF THE ACOUSTICAL SOCIETY OF AMERICA},

year = {2006},

volume = {119},

pages = {2498--2517}

}

### OpenURL

### Abstract

This paper deals with the computational analysis of musical audio from recorded audio waveforms. This general problem includes, as sub-tasks, music transcription, extraction of musical pitch, dynamics, timbre, instrument identity, and source separation. Analysis of real musical signals is a highly ill-posed task which is made complicated by the presence of transient sounds, background interference or the complex structure of musical pitches in the time-frequency domain. This paper focuses on models and algorithms for computer transcription of multiple musical pitches in audio, elaborated from previous work by two of the authors. The audio data are supposedly pre-segmented into fixed pitch regimes such as individual chords. The models presented apply to pitched (tonal) music and are formulated via a Gabor representation of non-stationary signals. A Bayesian probabilistic structure is employed for representation of prior information about the parameters of the notes. This paper introduces a numerical Bayesian inference strategy for estimation of the pitches and other parameters of the waveform. The improved algorithm is much quicker, and makes the approach feasible in realistic sitautions. Results are

### Citations

1840 |
A New Look at the Statistical Model Identification
- Akaike
- 1974
(Show Context)
Citation Context ..., see Rissanen[34] and Kameoka et al.[10]. This is a well-known over-fitting problem of ML procedures which can be dealt with in principle using over-fitting penalisation terms and methods such as AIC=-=[35]-=- and BIC[36] to determine an appropriate model order. Here we choose to penalise over-fitting of the data, both by the parameters β, σ 2 v, ω and the overall model order R in a fully Bayesian framewor... |

904 | Monte Carlo Statistical Methods
- ROBERT, G
(Show Context)
Citation Context ...r vector to be estimated. In this paper, θ is composed of β, σ 2 v, ω, M, K, and π(θ) = p(β, σ 2 v, ω, M, K|y). Two important algorithms are the Gibbs sampler and the Metropolis-Hastings sampler[41], =-=[42]-=-. MCMC algorithm have been widely used in Signal Processing applications[27], [28], [43], [44], [39], [45]. In this paper, we implement a specially constructed version of Metropolis-Hastings (MH) samp... |

826 | Reversible jump Markov chain Monte Carlo computation and Bayesian model determination
- Green
- 1995
(Show Context)
Citation Context .... This is not the case in our problem since the dimension of θ depends on M and K. Special extensions of MH algorithms, known as reversible jump MCMC, are designed to address cases where Nθ is unknown=-=[47]-=-, [42], [44], and we adopt these in this paper, following previous work[27], [28], [39], [29], [30]. B. Computation of parameter estimates We now assume that a Markov chain of samples has been generat... |

546 |
Markov Chain Monte Carlo in practice
- Gilks, Richardson, et al.
- 1996
(Show Context)
Citation Context ...romising here, either using informal ideas such as picking the chain with highest posterior probability or lowest residual error, or more formally using multiple chain MCMC methods and annealing, see =-=[49]-=-, [50]. These promising possibilities will be investigated in our next paper. Concerning extract #4.8, the error can be explained as follows: though this is not an octave error, we see that the third ... |

484 |
Auditory scene analysis
- Bregman
- 1990
(Show Context)
Citation Context ...nsidering the large quantities of prior information available about musical signals. Musical signals are highly structured, both in the time domain and in the frequency domain, see for example Bregman=-=[1]-=- or Fletcher and Rossing[2]. In the time domain, note transitions and percussive sounds occur at times related to the tempo and beat positions in the music. In the frequency domain, two levels of stru... |

398 |
A universal prior for integers and estimation by minimum description length
- Rissanen
- 1983
(Show Context)
Citation Context ...ers ( � β ML, �σv 2 ML, �ω ML, � M ML, � K ML), see Eq. (8). This approach is, however, not satisfactory in practice as it tends to favour solutions with too many partials Mk and notes K, see Rissanen=-=[34]-=- and Kameoka et al.[10]. This is a well-known over-fitting problem of ML procedures which can be dealt with in principle using over-fitting penalisation terms and methods such as AIC[35] and BIC[36] t... |

397 | Detection of Abrupt Changes: Theory and Application
- Basseville, Nikiforov
- 1993
(Show Context)
Citation Context ... note changes occur. Such segments can be obtained efficiently using segmentation algorithms based on timefrequency representations[21], Support Vector Machines[22], [23], Generalised Likelihood Ratio=-=[24]-=- or other more music-specific methods[25], [26]. We propose to describe these segments using a polyphonic music model where the number of partials, the time-varying amplitudes, the fundamental and ove... |

323 |
Estimating the dimension of a model
- Schwartz
- 1978
(Show Context)
Citation Context ...en[34] and Kameoka et al.[10]. This is a well-known over-fitting problem of ML procedures which can be dealt with in principle using over-fitting penalisation terms and methods such as AIC[35] and BIC=-=[36]-=- to determine an appropriate model order. Here we choose to penalise over-fitting of the data, both by the parameters β, σ 2 v, ω and the overall model order R in a fully Bayesian framework. This is a... |

109 | Dealing with label switching in mixture models
- Stephens
(Show Context)
Citation Context ...of β in our case (the ordering of individual notes in β, σ 2 v, ω and M is nonunique, this is the label switching problem). Bayesian estimation in label switching contexts is quite difficult. Stephens=-=[37]-=- proposes a relabelling algorithm, and Jasra[38] proposes a review of possible relabelling methods. The technique employed here is simpler – further discussion is left until the MCMC algorithms are pr... |

62 |
Digital audio restoration – A Statistical model-based approach
- Godsill, Rayner
- 1998
(Show Context)
Citation Context ...p(β, σ 2 v, ω, M, K|y). Two important algorithms are the Gibbs sampler and the Metropolis-Hastings sampler[41], [42]. MCMC algorithm have been widely used in Signal Processing applications[27], [28], =-=[43]-=-, [44], [39], [45]. In this paper, we implement a specially constructed version of Metropolis-Hastings (MH) sampling, whose principle is described in Algorithm 1 for a vector θ with a fixed number of ... |

59 |
Gabor analysis and algorithms. Theory and applications
- Feichtinger, Strohmer
- 1998
(Show Context)
Citation Context ...ave the frequency structure described above; however, the partials amplitudes are generally non-constant, see Fig. 1. An efficient modeling strategy is inspired by Gabor time-frequency representations=-=[31]-=- in which a signal y[t], t = 1, 2, . . . , N is projected on a basis of Gabor atoms well localised in time and frequency. Our polyphonic harmonic model relies on a family of realvalued and non-zero ph... |

58 | On the oscillations of musical instruments - McIntyre, Schumacher, et al. - 1983 |

54 |
Harmonic decomposition of audio signals with matching pursuit
- Gribonval, Bacry
- 2003
(Show Context)
Citation Context ... Gabor analysis, atom times {t0} and frequencies {ω0} are positioned on a regular time-frequency lattice – see Wolfe et al.[32] for a fully Bayesian analysis in this setting; see also Gribonval et al.=-=[33]-=- for Gabor atom-based analysis of musical audio within a projection pursuit setting. Here, a regular time lattice is kept, but the atom frequencies correspond to the partial frequencies ωk,m, and thes... |

53 |
Measuring the similarity of rhythmic patterns
- Paulus, Klapuri
- 2002
(Show Context)
Citation Context ... obtained efficiently using segmentation algorithms based on timefrequency representations[21], Support Vector Machines[22], [23], Generalised Likelihood Ratio[24] or other more music-specific methods=-=[25]-=-, [26]. We propose to describe these segments using a polyphonic music model where the number of partials, the time-varying amplitudes, the fundamental and overtone partial frequencies, and the noise ... |

45 | Bayesian harmonic models for musical signal analysis
- Davy, Godsill
- 2002
(Show Context)
Citation Context ...itudes, the fundamental and overtone partial frequencies, and the noise variances are unknown. This new model is slightly modified compared to the model presented in our previous work, see[27], [28], =-=[29]-=-, [30]. In particular, the noise is now assumed white (as opposed to colored in the previous Davy and Godsill papers), and the inharmonicity parameter takes a multiplicative form. These modifications ... |

42 |
Automatic Transcription of Piano Music
- Raphael
- 2002
(Show Context)
Citation Context ...for example de Cheveigné et al.[4], [5], Virtanen and Klapuri[6], [7], Irizarry[8], [9] or Kameoka[10]. Several existing Bayesian approaches include the work by Kashino et al.[11], [12], [13], Raphael=-=[14]-=-, [15], Tabrikian et al.[16], who adopt Bayesian hierarchical models for high level features in music such as chords, notes, timbre, etc., and Sterian et al.[17] who adopt Bayesian tracking ideas for ... |

42 | A Generative Model for Music Transcription
- Cemgil, Kappen, et al.
- 2006
(Show Context)
Citation Context ...ing of time varying frequency partials. In a related vein, Parra and Jain[18] present Kalman filtering estimators for the harmonic plus noise model. A recent development is that of Cemgil et al.[19], =-=[20]-=- who develop state-space models of musical harmonics and estimate them within an approximate Bayesian procedure. The complete polyphonic transcription task is a great technical challenge. In this pape... |

42 | An online kernel change detection algorithm
- Desobry, Davy, et al.
(Show Context)
Citation Context ...here it is assumed that no musical note changes occur. Such segments can be obtained efficiently using segmentation algorithms based on timefrequency representations[21], Support Vector Machines[22], =-=[23]-=-, Generalised Likelihood Ratio[24] or other more music-specific methods[25], [26]. We propose to describe these segments using a polyphonic music model where the number of partials, the time-varying a... |

35 | Detection of abrupt spectral changes using support vector machines an application to audio signal segmentation
- Davy, Godsill
- 2002
(Show Context)
Citation Context ...udio where it is assumed that no musical note changes occur. Such segments can be obtained efficiently using segmentation algorithms based on timefrequency representations[21], Support Vector Machines=-=[22]-=-, [23], Generalised Likelihood Ratio[24] or other more music-specific methods[25], [26]. We propose to describe these segments using a polyphonic music model where the number of partials, the time-var... |

35 | A hybrid approach to musical note onset detection
- Duxbury, Sandler, et al.
- 2002
(Show Context)
Citation Context ...ned efficiently using segmentation algorithms based on timefrequency representations[21], Support Vector Machines[22], [23], Generalised Likelihood Ratio[24] or other more music-specific methods[25], =-=[26]-=-. We propose to describe these segments using a polyphonic music model where the number of partials, the time-varying amplitudes, the fundamental and overtone partial frequencies, and the noise varian... |

34 | Polyphonic pitch tracking using joint bayesian estimation of multiple frame parameters
- Walmsley, Godsill, et al.
- 1999
(Show Context)
Citation Context ...g amplitudes, the fundamental and overtone partial frequencies, and the noise variances are unknown. This new model is slightly modified compared to the model presented in our previous work, see[27], =-=[28]-=-, [29], [30]. In particular, the noise is now assumed white (as opposed to colored in the previous Davy and Godsill papers), and the inharmonicity parameter takes a multiplicative form. These modifica... |

33 |
Sound source identification system for ensemble music based on template adaptation and music stream extraction,speech communication
- Kashino, Murase
- 1999
(Show Context)
Citation Context ...usical signal, see for example de Cheveigné et al.[4], [5], Virtanen and Klapuri[6], [7], Irizarry[8], [9] or Kameoka[10]. Several existing Bayesian approaches include the work by Kashino et al.[11], =-=[12]-=-, [13], Raphael[14], [15], Tabrikian et al.[16], who adopt Bayesian hierarchical models for high level features in music such as chords, notes, timbre, etc., and Sterian et al.[17] who adopt Bayesian ... |

31 | Multiple period estimation and pitch perception model
- Cheveigné, Kawahara
- 1999
(Show Context)
Citation Context ...n authors have adopted methods with statistical modelling elements, often using iterative procedures to estimate the individual components of a musical signal, see for example de Cheveigné et al.[4], =-=[5]-=-, Virtanen and Klapuri[6], [7], Irizarry[8], [9] or Kameoka[10]. Several existing Bayesian approaches include the work by Kashino et al.[11], [12], [13], Raphael[14], [15], Tabrikian et al.[16], who a... |

30 | On the Relationship Between Markov Chain Monte Carlo Methods for Model Uncertainty
- Godsill
- 2001
(Show Context)
Citation Context ... 2 v, ω, M, K|y). Two important algorithms are the Gibbs sampler and the Metropolis-Hastings sampler[41], [42]. MCMC algorithm have been widely used in Signal Processing applications[27], [28], [43], =-=[44]-=-, [39], [45]. In this paper, we implement a specially constructed version of Metropolis-Hastings (MH) sampling, whose principle is described in Algorithm 1 for a vector θ with a fixed number of dimens... |

29 | Harmonic analysis with probabilistic graphical models
- Raphael, Stoddard
- 2003
(Show Context)
Citation Context ...ample de Cheveigné et al.[4], [5], Virtanen and Klapuri[6], [7], Irizarry[8], [9] or Kameoka[10]. Several existing Bayesian approaches include the work by Kashino et al.[11], [12], [13], Raphael[14], =-=[15]-=-, Tabrikian et al.[16], who adopt Bayesian hierarchical models for high level features in music such as chords, notes, timbre, etc., and Sterian et al.[17] who adopt Bayesian tracking ideas for modell... |

26 | Bayesian harmonic models for musical pitch estimation and analysis
- Godsill, Davy
(Show Context)
Citation Context ...ns of MH algorithms, known as reversible jump MCMC, are designed to address cases where Nθ is unknown[47], [42], [44], and we adopt these in this paper, following previous work[27], [28], [39], [29], =-=[30]-=-. B. Computation of parameter estimates We now assume that a Markov chain of samples has been generated and that only the samples after convergence are kept. They are denoted ( � β (l) , �σ 2(l) v , �... |

23 |
Separation of harmonic sound sources using sinusoidal modeling
- Virtanen, Klapuri
- 2000
(Show Context)
Citation Context ...thods with statistical modelling elements, often using iterative procedures to estimate the individual components of a musical signal, see for example de Cheveigné et al.[4], [5], Virtanen and Klapuri=-=[6]-=-, [7], Irizarry[8], [9] or Kameoka[10]. Several existing Bayesian approaches include the work by Kashino et al.[11], [12], [13], Raphael[14], [15], Tabrikian et al.[16], who adopt Bayesian hierarchica... |

23 | Bayesian variable selection and regularisation for time–frequency surface estimation
- Wolfe, Godsill, et al.
- 2004
(Show Context)
Citation Context ...envelope with, e.g., Gaussian shape, Hamming shape, etc. In standard discrete Gabor analysis, atom times {t0} and frequencies {ω0} are positioned on a regular time-frequency lattice – see Wolfe et al.=-=[32]-=- for a fully Bayesian analysis in this setting; see also Gribonval et al.[33] for Gabor atom-based analysis of musical audio within a projection pursuit setting. Here, a regular time lattice is kept, ... |

17 | Generative model based polyphonic music transcription
- Cemgil, Kappen, et al.
- 2003
(Show Context)
Citation Context ...modelling of time varying frequency partials. In a related vein, Parra and Jain[18] present Kalman filtering estimators for the harmonic plus noise model. A recent development is that of Cemgil et al.=-=[19]-=-, [20] who develop state-space models of musical harmonics and estimate them within an approximate Bayesian procedure. The complete polyphonic transcription task is a great technical challenge. In thi... |

16 |
Stationarity index for abrupt changes detection in the time-frequency plane
- Laurent, Doncarli
- 1998
(Show Context)
Citation Context ...y estimation in segments of audio where it is assumed that no musical note changes occur. Such segments can be obtained efficiently using segmentation algorithms based on timefrequency representations=-=[21]-=-, Support Vector Machines[22], [23], Generalised Likelihood Ratio[24] or other more music-specific methods[25], [26]. We propose to describe these segments using a polyphonic music model where the num... |

15 |
Joint Bayesian Detection and Estimation of Noisy Sinusoids via Reversible Jump MCMC
- ANDRIEU, DOUCET
- 1999
(Show Context)
Citation Context ...o-mean Gaussian, with covariance matrix σ2 v ξ 2 I where I is an identity matrix and ξ 2 is a scaling parameter that can be interpreted as a signal to noise ratio, as pointed out by Andrieu and Doucet=-=[39]-=-. As explained in previous work[40], selecting a diagonal covariance matrix permits the fast MCMC implementation presented below. Moreover, this diagonal covariance matrix does not degrade noticeably ... |

15 | A.: On a class of genealogical and interacting metropolis models
- Moral, Doucet
- 2003
(Show Context)
Citation Context ...ng here, either using informal ideas such as picking the chain with highest posterior probability or lowest residual error, or more formally using multiple chain MCMC methods and annealing, see [49], =-=[50]-=-. These promising possibilities will be investigated in our next paper. Concerning extract #4.8, the error can be explained as follows: though this is not an octave error, we see that the third partia... |

12 |
Cheveigne, “Separation of concurrent harmonic sounds: Fundamental frequency estimation and a time-domain cancellation model of auditory processing
- de
- 1993
(Show Context)
Citation Context ...ertain authors have adopted methods with statistical modelling elements, often using iterative procedures to estimate the individual components of a musical signal, see for example de Cheveigné et al.=-=[4]-=-, [5], Virtanen and Klapuri[6], [7], Irizarry[8], [9] or Kameoka[10]. Several existing Bayesian approaches include the work by Kashino et al.[11], [12], [13], Raphael[14], [15], Tabrikian et al.[16], ... |

12 | Maximum a-posteriori probability pitch tracking in noisy environments using harmonic model
- Tabrikian, Dubnov, et al.
(Show Context)
Citation Context ...al.[4], [5], Virtanen and Klapuri[6], [7], Irizarry[8], [9] or Kameoka[10]. Several existing Bayesian approaches include the work by Kashino et al.[11], [12], [13], Raphael[14], [15], Tabrikian et al.=-=[16]-=-, who adopt Bayesian hierarchical models for high level features in music such as chords, notes, timbre, etc., and Sterian et al.[17] who adopt Bayesian tracking ideas for modelling of time varying fr... |

12 | Multidimensional optimisation of harmonic signals
- Walmsley, Godsill, et al.
- 1998
(Show Context)
Citation Context ...varying amplitudes, the fundamental and overtone partial frequencies, and the noise variances are unknown. This new model is slightly modified compared to the model presented in our previous work, see=-=[27]-=-, [28], [29], [30]. In particular, the noise is now assumed white (as opposed to colored in the previous Davy and Godsill papers), and the inharmonicity parameter takes a multiplicative form. These mo... |

11 |
Organisation of hierarchical perceptual sounds: music scene analysis with autonomous processing modules and a quantitative information integration mechanism
- KASHINO, NAKADAI, et al.
- 1995
(Show Context)
Citation Context ...of a musical signal, see for example de Cheveigné et al.[4], [5], Virtanen and Klapuri[6], [7], Irizarry[8], [9] or Kameoka[10]. Several existing Bayesian approaches include the work by Kashino et al.=-=[11]-=-, [12], [13], Raphael[14], [15], Tabrikian et al.[16], who adopt Bayesian hierarchical models for high level features in music such as chords, notes, timbre, etc., and Sterian et al.[17] who adopt Bay... |

9 | Separation of harmonic structures based on tied Gaussian mixture model and information criterion for concurrent sounds
- Kameoka, Nishimoto, et al.
- 2004
(Show Context)
Citation Context ...ents, often using iterative procedures to estimate the individual components of a musical signal, see for example de Cheveigné et al.[4], [5], Virtanen and Klapuri[6], [7], Irizarry[8], [9] or Kameoka=-=[10]-=-. Several existing Bayesian approaches include the work by Kashino et al.[11], [12], [13], Raphael[14], [15], Tabrikian et al.[16], who adopt Bayesian hierarchical models for high level features in mu... |

8 |
Approximate Kalman filtering for the harmonic plus noise model
- Para, Jain
- 2001
(Show Context)
Citation Context ...evel features in music such as chords, notes, timbre, etc., and Sterian et al.[17] who adopt Bayesian tracking ideas for modelling of time varying frequency partials. In a related vein, Parra and Jain=-=[18]-=- present Kalman filtering estimators for the harmonic plus noise model. A recent development is that of Cemgil et al.[19], [20] who develop state-space models of musical harmonics and estimate them wi... |

8 |
Bayesian model selection applied to self-noise geoacoustic inversion
- Battle, Gerstoft, et al.
(Show Context)
Citation Context ...escribed in Algorithm 1 for a vector θ with a fixed number of dimensions Nθ and for the target distribution π(θ). Another presentation of MCMC in the context of Acoustics can be found in Battle et al.=-=[46]-=-. Algorithm 1 consists of sampling the components of θ one at a time, according to a distribution q(θ). This distribution is called a proposal, because it is used to form the proposed candidate parame... |

6 | Local harmonic estimation in musical sound signals
- Irizarry
- 2001
(Show Context)
Citation Context ...ical modelling elements, often using iterative procedures to estimate the individual components of a musical signal, see for example de Cheveigné et al.[4], [5], Virtanen and Klapuri[6], [7], Irizarry=-=[8]-=-, [9] or Kameoka[10]. Several existing Bayesian approaches include the work by Kashino et al.[11], [12], [13], Raphael[14], [15], Tabrikian et al.[16], who adopt Bayesian hierarchical models for high ... |

6 |
Bayesian estimation of simultaneous musical notes based on frequency domain modelling
- Kashino, Godsill
- 2004
(Show Context)
Citation Context ... signal, see for example de Cheveigné et al.[4], [5], Virtanen and Klapuri[6], [7], Irizarry[8], [9] or Kameoka[10]. Several existing Bayesian approaches include the work by Kashino et al.[11], [12], =-=[13]-=-, Raphael[14], [15], Tabrikian et al.[16], who adopt Bayesian hierarchical models for high level features in music such as chords, notes, timbre, etc., and Sterian et al.[17] who adopt Bayesian tracki... |

4 | Fast MCMC computations for the estimation of sparse processes from noisy observations
- Davy, Idier
- 2004
(Show Context)
Citation Context ...is dramatically quicker than that of D t D (matrix inversion algorithms complexity scales as N 3 mat, where N mat is the matrix dimension). More details about this trick can be found in Davy and Idier=-=[40]-=-, where several fast methods are also proposed to compute r t aPara, det(Sa), µ a and to sample from N (µ a, σ 2 vSa) based on the Choleski decomposition of S −1 a (where the notation with subscript a... |

3 |
estimation of harmonic components in a musical sound signal
- “Weighted
(Show Context)
Citation Context ...modelling elements, often using iterative procedures to estimate the individual components of a musical signal, see for example de Cheveigné et al.[4], [5], Virtanen and Klapuri[6], [7], Irizarry[8], =-=[9]-=- or Kameoka[10]. Several existing Bayesian approaches include the work by Kashino et al.[11], [12], [13], Raphael[14], [15], Tabrikian et al.[16], who adopt Bayesian hierarchical models for high level... |

3 |
Modelbased musical transcription
- Sterian, Simoni, et al.
- 1999
(Show Context)
Citation Context ...Kashino et al.[11], [12], [13], Raphael[14], [15], Tabrikian et al.[16], who adopt Bayesian hierarchical models for high level features in music such as chords, notes, timbre, etc., and Sterian et al.=-=[17]-=- who adopt Bayesian tracking ideas for modelling of time varying frequency partials. In a related vein, Parra and Jain[18] present Kalman filtering estimators for the harmonic plus noise model. A rece... |

1 | MCMC and the label switching problem in Bayesian mixture models
- Jasra, Holmes, et al.
- 2005
(Show Context)
Citation Context ...es in β, σ 2 v, ω and M is nonunique, this is the label switching problem). Bayesian estimation in label switching contexts is quite difficult. Stephens[37] proposes a relabelling algorithm, and Jasra=-=[38]-=- proposes a review of possible relabelling methods. The technique employed here is simpler – further discussion is left until the MCMC algorithms are presented. March 30, 2006 DRAFT 6 (8)sThe posterio... |

1 | Reversible jump Markov chain Monte Carlo strategies for Bayesian model selection in autoregressive processes
- Vermaak, Andrieu, et al.
- 2004
(Show Context)
Citation Context ...K|y). Two important algorithms are the Gibbs sampler and the Metropolis-Hastings sampler[41], [42]. MCMC algorithm have been widely used in Signal Processing applications[27], [28], [43], [44], [39], =-=[45]-=-. In this paper, we implement a specially constructed version of Metropolis-Hastings (MH) sampling, whose principle is described in Algorithm 1 for a vector θ with a fixed number of dimensions Nθ and ... |