An utterance comparison model for speaker clustering using factor analysis

Jeon, Woojay; Ma, Changxue; Macho, Dušan

doi:10.1109/icassp.2011.5947361

Cited by 6 publications

(5 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Another method is to simply let and in (55) include these dropped utterances while ignoring the dropped utterances in the computation of , hence reflecting the errors in . In this work, we chose the latter method (in our previous work [20], we used the former method, which is why the absolute performance figures differ; also note that in this work, we used the MACROPHONE corpus for training, which is also different).…”

Section: Speaker Clustering Experimentsmentioning

confidence: 96%

“…In Section III, a closed-form solution of the model is derived by incorporating speaker factors. We include additional details to more clearly explain the derivation that was done in [20]. Next, in Section IV, the solution is extended by adding channel factors, which results in a final model that is identical in form to the initial model but with an alternate set of sufficient statistics that is inclusive of the initial set of statistics.…”

Section: Introductionmentioning

confidence: 98%

See 1 more Smart Citation

Statistical Utterance Comparison for Speaker Clustering Using Factor Analysis

Jeon

Ma²,

Macho

2012

IEEE Trans. Audio Speech Lang. Process.

Self Cite

View full text Add to dashboard Cite

We propose a novel method of measuring the similarity between two or more speech utterances for speaker clustering, based on probability theory and factor analysis. The similarity function is formulated as the probability that the utterances originated from the same speaker, and uses statistical eigenvoice and eigenchannel models to incorporate physical knowledge of interspeaker and intraspeaker variabilities, allowing the similarity function to be trainable and robust. The comparison function can be efficiently computed using a compact set of sufficient statistics for each speech utterance, allowing the acoustic features to be discarded. We begin using only eigenvoices, and then show how the eigenchannels can be incorporated into the equation to result in an identical form but with a different set of sufficient statistics. We test the proposed model in a speaker clustering task using the CALLHOME telephone conversation corpus and show that it performs better than two other well-known similarity measures: the Cross-Likelihood Ratio (CLR) and Generalized Likelihood Ratio (GLR).

show abstract

Section: Speaker Clustering Experimentsmentioning

confidence: 96%

Section: Introductionmentioning

confidence: 98%

Statistical Utterance Comparison for Speaker Clustering Using Factor Analysis

Jeon

Ma²,

Macho

2012

IEEE Trans. Audio Speech Lang. Process.

Self Cite

View full text Add to dashboard Cite

show abstract

“…Statistical methods such as Bayesian Information Criterion (e.g., [ 92 ]), Generalized Likelihood Ratio (e.g., [ 93 ]), Kullback-Leibler divergence (e.g., [ 94 ]), and Information Change Rate (e.g., [ 95 ]) are used for both segmenting and clustering the speech. Lastly, machine learning methods such as Variational Bayes (e.g., [ 96 ]), Non-parametric Bayes (e.g., [ 97 ]), and Factor Analysis (e.g., [ 98 ]) are introduced to solve the diarization problem.…”

Section: Figure A1mentioning

confidence: 99%

The Promise and the Challenge of Technology-Facilitated Methods for Assessing Behavioral and Cognitive Markers of Risk for Suicide among U.S. Army National Guard Personnel

Baucom

Georgiou

Bryan

et al. 2017

IJERPH

View full text Add to dashboard Cite

Suicide was the 10th leading cause of death for Americans in 2015 and rates have been steadily climbing over the last 25 years. Rates are particularly high amongst U.S. military personnel. Suicide prevention efforts in the military are significantly hampered by the lack of: (1) assessment tools for measuring baseline risk and (2) methods to detect periods of particularly heightened risk. Two specific barriers to assessing suicide risk in military personnel that call for innovation are: (1) the geographic dispersion of military personnel from healthcare settings, particularly amongst components like the Reserves; and (2) professional and social disincentives to acknowledging psychological distress. The primary aim of this paper is to describe recent technological developments that could contribute to risk assessment tools that are not subject to the limitations mentioned above. More specifically, Behavioral Signal Processing can be used to assess behaviors during interaction and conversation that likely indicate increased risk for suicide, and computer-administered, cognitive performance tasks can be used to assess activation of the suicidal mode. These novel methods can be used remotely and do not require direct disclosure or endorsement of psychological distress, solving two challenges to suicide risk assessment in military and other sensitive settings. We present an introduction to these technologies, describe how they can specifically be applied to assessing behavioral and cognitive risk for suicide, and close with recommendations for future research.

show abstract

“…Speaker clustering is an effective tool that can alleviate the amount of speech document management tasks [5][6]. Speaker clustering can group similar audio utterance together and attribute it to the same speaker in audio document by some distance measure and clustering scheme in the unsupervised condition [7].…”

Section: Introductionmentioning

confidence: 99%

An Algorithm of Speaker Clustering Based on Model Distance

Yang

2014

JMM

View full text Add to dashboard Cite

An algorithm based on Model Distance (MD) for spectral speaker clustering is proposed to deal with the shortcoming of general spectral clustering algorithm in describing the distribution of signal source. First, an Universal Background Model (UBM) is created with a large quantity of independent speakers; Then, Gaussian Mixture Model (GMM) is trained from the UBM for every speech segment; At last, the probability distance between the GMM of every speech segment is used to build affinity matrix, and speaker spectral clustering is done on the affinity matrix. Experimental results based on news and conference data sets show that an average of 6.38% improvements in F measure is obtained in comparison with algorithm based on the feature vector distance. In addition, the proposed algorithm is 11.72 times faster.

show abstract

An utterance comparison model for speaker clustering using factor analysis

Cited by 6 publications

References 7 publications

Statistical Utterance Comparison for Speaker Clustering Using Factor Analysis

Statistical Utterance Comparison for Speaker Clustering Using Factor Analysis

The Promise and the Challenge of Technology-Facilitated Methods for Assessing Behavioral and Cognitive Markers of Risk for Suicide among U.S. Army National Guard Personnel

An Algorithm of Speaker Clustering Based on Model Distance

Contact Info

Product

Resources

About