Increasing the Robustness of i-vectors with Model Compensated First Order Statistics

Gökay Dişken; Zekeriya Tüfekci

Article Details

Increasing the Robustness of i-vectors with Model Compensated First Order Statistics

Authors:

Gökay Dişken | Zekeriya Tüfekci

Publisher:

Afyon Kocatepe University

Main Topics:

Elektrik Mühendisliği

Journal:

Afyon Kocatepe Üniversitesi Fen Ve Mühendislik Bilimleri Dergisi Vol.23 Issue 1, pp.123-137

Document Type:

Article

Publication Date:

1/2023

Document Language:

English

Document ID (Org No.):

2001284425

pISSN:

2149-3367

Keywords:

Abstract:

 Speaker recognition systems achieved significant improvements over the last decade, especially due to the performance of the i-vectors. Despite the achievements, mismatch between training and test data affects the recognition performance considerably. In this paper, a solution is offered to increase robustness against additive noises by inserting model compensation techniques within the i-vector extraction scheme. For stationary noises, the model compensation techniques produce highly robust systems. Parallel Model Compensation and Vector Taylor Series are considered as state-of-the-art model compensation techniques. Applying these methods to the first order statistics, a noisy total variability space training is aimed, which will reduce the mismatch resulted by additive noises. All other parts of the conventional i-vector scheme remain unchanged, such as total variability matrix training, reducing the i-vector dimensionality, scoring the i-vectors. The proposed method was tested with four different noise types with several signal to noise ratios SNR from -6 dB to 18 dB with 6 dB steps. High reductions in equal error rates were achieved with both methods, even at the lowest SNR levels. On average, the proposed approach produced more than 50 relative reduction in equal error rate.

 Speaker recognition systems achieved significant improvements over the last decade, especially due to the performance of the i-vectors. Despite the achievements, mismatch between training and test data affects the recognition performance considerably. In this paper, a solution is offered to increase robustness against additive noises by inserting model compensation techniques within the i-vector extraction scheme. For stationary noises, the model compensation techniques produce highly robust systems. Parallel Model Compensation and Vector Taylor Series are considered as state-of-the-art model compensation techniques. Applying these methods to the first order statistics, a noisy total variability space training is aimed, which will reduce the mismatch resulted by additive noises. All other parts of the conventional i-vector scheme remain unchanged, such as total variability matrix training, reducing the i-vector dimensionality, scoring the i-vectors. The proposed method was tested with four different noise types with several signal to noise ratios SNR from -6 dB to 18 dB with 6 dB steps. High reductions in equal error rates were achieved with both methods, even at the lowest SNR levels. On average, the proposed approach produced more than 50 relative reduction in equal error rate.

 Speaker recognition systems achieved significant improvements over the last decade, especially due to the performance of the i-vectors. Despite the achievements, mismatch between training and test data affects the recognition performance considerably. In this paper, a solution is offered to increase robustness against additive noises by inserting model compensation techniques within the i-vector extraction scheme. For stationary noises, the model compensation techniques produce highly robust systems. Parallel Model Compensation and Vector Taylor Series are considered as state-of-the-art model compensation techniques. Applying these methods to the first order statistics, a noisy total variability space training is aimed, which will reduce the mismatch resulted by additive noises. All other parts of the conventional i-vector scheme remain unchanged, such as total variability matrix training, reducing the i-vector dimensionality, scoring the i-vectors. The proposed method was tested with four different noise types with several signal to noise ratios SNR from -6 dB to 18 dB with 6 dB steps. High reductions in equal error rates were achieved with both methods, even at the lowest SNR levels. On average, the proposed approach produced more than 50 relative reduction in equal error rate.

 Konuşmacı tanıma sistemleri özellikle i-vektörlerin performansı sebebiyle son on yılda önemli gelişmeler elde etmiştir. Bu gelişmelere rağmen eğitim ve test verileri arasındaki uyumsuzluk tanıma performansını önemli ölçüde etkilemektedir. Bu çalışmada, model kompanzasyon yöntemleri i-vektör çıkarımı şemasına eklenerek toplanabilir gürültülere karşı gürbüzlüğü artıracak bir çözüm sunulmaktadır. Durağan gürültüler için model kompanzasyon teknikleri oldukça gürbüz sistemler üretir. Paralel Model Kompanzasyonu ve Vektör Taylor Serileri en gelişmiş model kompanzasyon tekniklerinden kabul edilmektedir. Bu metotlar birinci dereceden istatistiklere uygulanarak toplanabilir gürültülerden kaynaklanan uyumsuzluğu azaltacak gürültülü tüm değişkenlik uzayı eğitimi amaçlanmıştır. Tüm değişkenlik matrisin eğitimi, i-vektör boyutunun azaltılması, i-vektörlerin puanlanması gibi geleneksel i-vektör şemasının diğer tüm parçaları değişmeden kalmaktadır. Önerilen yöntem, 6 dBlik adımlarla -6 dBden 18 dBye kadar çeşitli sinyal-gürültü oranlarına SNR sahip dört farklı gürültü tipi ile test edilmiştir. Her iki yöntemle de en düşük SNR seviyelerinde bile eşit hata oranlarında yüksek azalmalar elde edilmiştir. Önerilen yaklaşım eşik hata oranında ortalama olarak 50den fazla göreceli azalma sağlamıştır.

Article Details

Increasing the Robustness of i-vectors with Model Compensated First Order Statistics

Authors:

Gökay Dişken | Zekeriya Tüfekci

Publisher:

Afyon Kocatepe University

Main Topics:

Elektrik Mühendisliği

Journal:

Afyon Kocatepe Üniversitesi Fen Ve Mühendislik Bilimleri Dergisi Vol.23 Issue 1, pp.123-137

Document Type:

Article

Publication Date:

1/2023

Document Language:

English

Document ID (Org No.):

2001284425

pISSN:

2149-3367

Parallel model compensation

Robust speaker recognition

Vector Taylor series

I-vector

Paralel model kompanzasyonu

Gürbüz konuşmacı tanıma

Vektör Taylor serileri

I-vektör

Issues

Article Details

Authors: Gökay Dişken | Zekeriya Tüfekci

Publisher: Afyon Kocatepe University

Main Topics: Elektrik Mühendisliği

Journal: Afyon Kocatepe Üniversitesi Fen Ve Mühendislik Bilimleri Dergisi Vol.23 Issue 1, pp.123-137

Document Type: Article

Publication Date: 1/2023

Document Language: English

Document ID (Org No.): 2001284425

pISSN: 2149-3367

Issues

Authors:

Gökay Dişken | Zekeriya Tüfekci

Publisher:

Afyon Kocatepe University

Main Topics:

Elektrik Mühendisliği

Journal:

Afyon Kocatepe Üniversitesi Fen Ve Mühendislik Bilimleri Dergisi Vol.23 Issue 1, pp.123-137

Document Type:

Article

Publication Date:

1/2023

Document Language:

English

Document ID (Org No.):

2001284425

pISSN:

2149-3367