Generalized Viterbi-based models for time-series segmentation and clustering applied to speaker diarization

Lapidot, Itshak; Shoa, Alon; Furmanov, Tal; Aminov, Lidiya; Moyal, Ami; Bonastre, Jean-François

doi:10.1016/j.csl.2017.01.011

Cited by 10 publications

(11 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…There are 109, 9 and 2 conversations containing 2, 3 and 4 speakers, respectively. We follow the practice of [55] and [56], conversations with 2 speakers are examined. We use Switchboard P1-3/Cell and SRE04-06 to train UBM, T and PLDA parameters.…”

Section: Experiments Results With Callhome97mentioning

confidence: 99%

Latent class model with application to speaker diarization

Chen

et al. 2019

J AUDIO SPEECH MUSIC PROC.

View full text Add to dashboard Cite

In this paper, we apply a latent class model (LCM) to the task of speaker diarization. LCM is similar to Patrick Kenny's variational Bayes (VB) method in that it uses soft information and avoids premature hard decisions in its iterations. In contrast to the VB method, which is based on a generative model, LCM provides a framework allowing both generative and discriminative models. The discriminative property is realized through the use of i-vector (Ivec), probabilistic linear discriminative analysis (PLDA), and a support vector machine (SVM) in this work. Systems denoted as LCM-Ivec-PLDA, LCM-Ivec-SVM, and LCM-Ivec-Hybrid are introduced. In addition, three further improvements are applied to enhance its performance. 1) Adding neighbor windows to extract more speaker information for each short segment. 2) Using a hidden Markov model to avoid frequent speaker change points. 3) Using an agglomerative hierarchical cluster to do initialization and present hard and soft priors, in order to overcome the problem of initial sensitivity. Experiments on the National Institute of Standards and Technology Rich Transcription 2009 speaker diarization database, under the condition of a single distant microphone, show that the diarization error rate (DER) of the proposed methods has substantial relative improvements compared with mainstream systems. Compared to the VB method, the relative improvements of LCM-Ivec-PLDA, LCM-Ivec-SVM, and LCM-Ivec-Hybrid systems are 23.5%, 27.1%, and 43.0%, respectively. Experiments on our collected database, CALLHOME97, CALLHOME00 and SRE08 short2-summed trial conditions also show that the proposed LCM-Ivec-Hybrid system has the best overall performance.

show abstract

Section: Experiments Results With Callhome97mentioning

confidence: 99%

Latent class model with application to speaker diarization

Chen

et al. 2019

J AUDIO SPEECH MUSIC PROC.

View full text Add to dashboard Cite

show abstract

“…To confirm this fact, we display in Figure 3 information similar to that in Figure 1 but with PMFs computed only on speech parts when Figure 4 proposes the PMFs of the nonspeech parts (both experiments are using the same VAD process). We apply a very simple energy VAD, using the same approach as in [31] and [32]. When there is almost no difference between genuine and spoofing speech PMFs for the speech part, a significant difference is observed for the non-speech part.…”

Section: Pmfs Of Genuine and Spoofing Speechmentioning

confidence: 99%

Effects of Waveform PMF on Anti-Spoofing Detection

Lapidot

Bonastre

2019

Interspeech 2019

Self Cite

View full text Add to dashboard Cite

In the context of detection of speaker recognition identity impersonation, we observed that the waveform probability mass function (PMF) of genuine speech differs from significantly of of PMF from identity theft extracts. This is true for synthesized or converted speech as well as for replayed speech. In this work, we mainly ask whether this observation has a significant impact on spoofing detection performance. In a second step, we want to reduce the distribution gap of waveforms between authentic speech and spoofing speech. We propose a genuinization of the spoofing speech (by analogy with Gaussianisation), i.e. to obtain spoofing speech with a PMF close to the PMF of genuine speech. Our genuinization is evaluated on ASVspoof 2019 challenge datasets, using the baseline system provided by the challenge organization. In the case of constant Q cepstral coefficients (CQCC) features, the genuinization leads to a degradation of the baseline system performance by a factor of 10, which shows a potentially large impact of the distribution os waveforms on spoofing detection performance. However, by ''playing" with all configurations, we also observed different behaviors, including performance improvements in specific cases. This leads us to conclude that waveform distribution plays an important role and must be taken into account by antispoofing systems.

show abstract

“…The histogram of waveform amplitude is determined by dividing the bin counts by the number of samples in the signal, thereby giving the signal amplitude probability mass function (PMF). The entropy of the signal amplitude PMF is then determined according to the standard approach given by: 16 (1)…”

Section: Entropy Validationmentioning

confidence: 99%

“…Accordingly, VAD was applied to TIMIT and RSR datasets, but not to the ASVspoof datasets. VAD is performed according to approach described in [15] and [16]. Every 100ms a norm-2 signal was calculated according to En = 1599 k=0 s 2 (1600n + k) 0.5 .…”

Section: Pre-processingmentioning

confidence: 99%

Speech Database and Protocol Validation Using Waveform Entropy

et al. 2018

Self Cite

View full text Add to dashboard Cite

The assessment of performance for any number of speech processing tasks calls for the use of a suitably large, representative dataset. Dataset design is crucial so as to ensure that any significant variation unrelated to the task in hand is adequately normalised or marginalised. Most datasets are partitioned into training, development and evaluation subsets. Depending on the task, the nature of these three subsets should normally be close to identical. With speech signals being subject to a multitude of different influences, e.g. speaker gender and age, language, dialect, utterance length, etc., the design and validation of speech datasets can become especially challenging. Even if many sources of variation unrelated to the task in hand can easily be marginalised, other sources of more subtle variation can easily be overlooked. Imbalances between training, development and evaluation partitions, can bring into question findings derived from their use. Stringent dataset validation procedures are required. This paper reports a particularly straightforward approach to dataset validation that is based upon waveform entropy.

show abstract

Generalized Viterbi-based models for time-series segmentation and clustering applied to speaker diarization

Cited by 10 publications

References 8 publications

Latent class model with application to speaker diarization

Latent class model with application to speaker diarization

Effects of Waveform PMF on Anti-Spoofing Detection

Speech Database and Protocol Validation Using Waveform Entropy

Contact Info

Product

Resources

About