Deep Learning Backend for Single and Multisession i-Vector Speaker Recognition

Ghahabi, Omid; Hernando, Javier

doi:10.1109/taslp.2017.2661705

Cited by 55 publications

(31 citation statements)

References 42 publications

(69 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…All the speech segments are chopped into 2 sec segments and converted to 300 dimensional speaker vectors. Afterwards, the resulting speaker vectors are clustered using a two stage unsupervised clustering technique which was used to estimate the speaker labels of the background data for training the Probabilistic Linear Discriminant Analysis (PLDA) in [4]. The first stage of the clustering algorithm is similar to the Mean Shift based algorithm proposed in [5] and used successfully in [6].…”

Section: Scoringmentioning

confidence: 99%

EML Submission to Albayzin 2018 Speaker Diarization Challenge

Ghahabi¹,

Fischer²

2018

IberSPEECH 2018

Self Cite

View full text Add to dashboard Cite

Speaker diarization, who is speaking when, is one of the most challenging tasks in speaker recognition, as usually no prior information is available about the identity and the number of the speakers in an audio recording. The task will be more challenging when there is some noise or music on the background and the speakers are changed more frequently. This usually happens in broadcast news conversations. In this paper, we use the EML speaker diarization system as a participation to the recent Albayzin Evaluation challenge. The EML system uses a realtime robust algorithm to make decision about the identity of the speakers approximately every 2 sec. The experimental results on about 16 hours of the developing data provided in the challenge show a reasonable accuracy of the system with a very low computational cost.

show abstract

Section: Scoringmentioning

confidence: 99%

EML Submission to Albayzin 2018 Speaker Diarization Challenge

Ghahabi¹,

Fischer²

2018

IberSPEECH 2018

Self Cite

View full text Add to dashboard Cite

show abstract

“…The main research trend consists of designing biometric recognition methods that are robust to poor-quality signals, and the research community is mainly focused on DL techniques, which learn the discriminative representation of an individual directly from the raw input signal [18].…”

Section: Voicementioning

confidence: 99%

“…Recently, biometric systems based on DL techniques and CNNs have been gaining popularity and have achieved accuracy improvements for face, fingerprint [30], iris [6], palm [49], ECG [44], voice [18], and gait [50] recognition, as well as for age and gender estimation. DL techniques are also being used in multibiometric systems to increase accuracy [2] or to learn multiple representations from the same biometric sample [22].…”

Section: Accuracy and Execution Timementioning

confidence: 99%

Advanced Biometric Technologies: Emerging Scenarios and Research Trends

Genovese

Muñoz

Piuri

et al. 2018

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Biometric systems are the ensemble of devices, procedures, and algorithms for the automatic recognition of individuals by means of their physiological or behavioral characteristics. Although biometric systems are traditionally used in high-security applications, recent advancements are enabling the application of these systems in less-constrained conditions with non-ideal samples and with real-time performance. Consequently, biometric technologies are being increasingly used in a wide variety of emerging application scenarios, including public infrastructures, e-government, humanitarian services, and user-centric applications. This chapter introduces recent biometric technologies, reviews emerging scenarios for biometric recognition, and discusses research trends.

show abstract

“…As an example of frontend, in [22,25], a vector representation of speakers was proposed by means of RBM adaptation. As a backend, in [26], various imposter selection algorithms are proposed, in order to reduce the performance gap between cosine and PLDA scoring techniques without the use of labeled data. They applied DBN adaptation as a backend for i-vector based speaker verification.…”

Section: Introductionmentioning

confidence: 99%

Auto-Encoding Nearest Neighbor i-Vectors for Speaker Verification

Khan¹,

India²,

Hernando³

2019

Interspeech 2019

Self Cite

View full text Add to dashboard Cite

In the last years, i-vectors followed by cosine or PLDA scoring techniques were the state-of-the-art approach in speaker verification. PLDA requires labeled background data, and there exists a significant performance gap between the two scoring techniques. In this work, we propose to reduce this gap by using an autoencoder to transform i-vector into a new speaker vector representation, which will be referred to as ae-vector. The autoencoder will be trained to reconstruct neighbor i-vectors instead of the same training i-vectors, as usual. These neighbor i-vectors will be selected in an unsupervised manner according to the highest cosine scores to the training i-vectors. The evaluation is performed on the speaker verification trials of VoxCeleb-1 database. The experiments show that our proposed ae-vectors gain a relative improvement of 42% in terms of EER compared to the conventional i-vectors using cosine scoring, which fills the performance gap between cosine and PLDA scoring techniques by 92%, but without using speaker labels.

show abstract

Deep Learning Backend for Single and Multisession i-Vector Speaker Recognition

Cited by 55 publications

References 42 publications

EML Submission to Albayzin 2018 Speaker Diarization Challenge

EML Submission to Albayzin 2018 Speaker Diarization Challenge

Advanced Biometric Technologies: Emerging Scenarios and Research Trends

Auto-Encoding Nearest Neighbor i-Vectors for Speaker Verification

Contact Info

Product

Resources

About