Model Complexity Selection and Cross-Validation EM Training for Robust Speaker Diarization

Anguera, Xavier; Shinozaki, Takahiro; Woofers, C.; Hernando, Javier

doi:10.1109/icassp.2007.366902

Cited by 10 publications

(3 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Table 5 These results suggest that the CL level has a lower influence on the phone segments than on the pause segments. [31] using the Pa-2010 feature set. Table 6 gives the UAR on the Development set of the C2 and C2N classifiers for the three CLSE databases.…”

Section: Audio Analysis Of Pause and Phone Segmentsmentioning

confidence: 99%

High-level speech event analysis for cognitive load classification

Montacié¹,

Caraty

2014

Interspeech 2014

View full text Add to dashboard Cite

The Cognitive Load (CL) refers to the load imposed on an individual's cognitive system when performing a given task, and is usually associated with the limitations of the human working memory. Stress, fatigue, lower ability to make decisions and perceptual narrowing are induced by cognitive overload which occurs when too much information has to be processed. As many physiological measures and for a nonintrusive measurement, speech features have been investigated in order to find reliable indicators of CL levels. In this paper, we have investigated high-level speech events automatically detected using the CMU-Sphinx toolkit for speech recognition. Temporal events (speech onset latency, event starting timecodes, pause and phone segments) were extracted from the speech transcriptions (phoneme, word, silent pause, filled pause, breathing). Seven audio feature sets related to the speech events were designed and assessed. Three-class SVM classifiers (Low, Medium and High level) were developed and assessed on the CSLE (Cognitive-Load with Speech and EGG) databases provided for the Interspeech'2014 Cognitive Load Sub-Challenge. These experiments have shown an improvement of 1.5 % on the Test set compared to the official baseline Unweighted Average Recall (UAR).

show abstract

Section: Audio Analysis Of Pause and Phone Segmentsmentioning

confidence: 99%

High-level speech event analysis for cognitive load classification

Montacié¹,

Caraty

2014

Interspeech 2014

View full text Add to dashboard Cite

show abstract

“…Here, for speaker change detection αδ is regarded as the threshold. The main goal of the validation approach [11][12] is verification of speaker change point detected in Step 3. From the graph, P and Q are two peak points of the correlation coefficient curve between U, V, and W that are 3 adjoining valleys in Fig.…”

Section: B Segmentation Algorithmmentioning

confidence: 99%

A new speaker change detection method in a speaker identification system for two-speakers segmentation

Bazyar

Sudirman

2014

2014 IEEE Symposium on Computer Applications and Industrial Electronics (ISCAIE)

View full text Add to dashboard Cite

Speaker change detection is done in many speaker and speech identification applications that the speech is from two speakers. However, the standard metric-based methods performance is not suitable and stable owing to the amid window distance calculation stability. Therefore, a new method is proposed to improve the stability and enhance the performance of the system according to speakers' characteristics using between window correlations. Moreover, reference speaker models set that shows the space of the entire speaker model are trained in this approach. A metric is defined as the between window correlation of scores likelihood vectors versus the reference models. The Peak and Valley information and gender information are also used. In this paper, we look at telephone conversations where it is known a priori that there are two speakers, but the identity of the speakers is not known. Experiments over Farsdat Database show better performance In comparison with the GLR and the BIC approaches. This new approach has more effect rather than the GLR and the BIC approaches in the broader value of defined thresholds.

show abstract

“…Each state of an acoustic model contains several Gaussian mixtures. According to [14], a parametric modeling technique, model complexity selection, is chosen to perform on each state with sufficient training data, and we select the number of mixtures based on the number of data frames belonging to the phoneme state in the acoustic model. Model complexity selection works as follows: whenever there is a change in the amount of data assigned to a model, the number of the available training samples that are assigned to the model is used to determine the new number of mixtures in the GMM using: …”

Section: Model Complexity Selection (Mcs)mentioning

confidence: 99%

Optimizing the acoustic modeling from an unbalanced bi-lingual corpus

Lyu

2008

2008 IEEE International Conference on Acoustics, Speech and Signal Processing

View full text Add to dashboard Cite

Phoneme set clustering of accurate modeling is important in the task of multilingual speech recognition, especially when each of the available language training corpora is mismatched, such as is the case between a major language, like Mandarin, and a minor language, like Taiwanese. In this paper, we present a data-driven approach for not only acquiring a proper phoneme set but optimizing the acoustic modeling in this situation. In order to obtain the phoneme set that is suitable for the unbalanced corpus, we use an agglomerative hierarchical clustering with delta Bayesian information criteria. Then for training each of the acoustic models, we choose a parametric modeling technique, model complexity selection, to adjust the number of mixtures for optimizing the acoustic model between the new phoneme set and the available training data. The experimental results are very encouraging in that the proposed approach reduces relative syllable error rate by 7.8% over the best result of the knowledge-based approach.

show abstract

Model Complexity Selection and Cross-Validation EM Training for Robust Speaker Diarization

Cited by 10 publications

References 8 publications

High-level speech event analysis for cognitive load classification

High-level speech event analysis for cognitive load classification

A new speaker change detection method in a speaker identification system for two-speakers segmentation

Optimizing the acoustic modeling from an unbalanced bi-lingual corpus

Contact Info

Product

Resources

About