Optimizing acoustic features for source cell-phone recognition using speech signals

Hanilçi, Cemal; Ertaş, Figen

doi:10.1145/2482513.2482520

Cited by 17 publications

(9 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We compare the source identification performances of these two different 100 acoustic feature sets. Detailed comparison of different feature extraction and normalization methods on source cell-phone identification can be found in [18].…”

mentioning

confidence: 99%

Source cell-phone recognition from recorded speech using non-speech segments

Hanilçi

Kinnunen

2014

Digital Signal Processing

Self Cite

View full text Add to dashboard Cite

In a recent study, we have introduced the problem of identifying cell-phones using recorded speech and shown that speech signals convey information about the source device, making it possible to identify the source with some accuracy. In this paper, we consider recognizing source cell-phone microphones using non-speech segments of recorded speech. Taking an information-theoretic approach, we use Gaussian Mixture Model (GMM) trained with maximum mutual information (MMI) to represent device-specific features. Experimental results using Mel-frequency and linear frequency cepstral coefficients (MFCC and LFCC) show that features extracted from the non-speech segments of speech contain higher mutual information and yields higher recognition rates than those from speech portions or the whole utterance. Identification rate improves from 96.42 % to 98.39 % and equal error rate (EER) reduces from 1.20% to 0.47 % when non-speech parts are used to extract features. Recognition results are provided with classical GMM trained both with maximum likelihood (ML) and maximum mutual information (MMI) criteria, as well as support vector machines (SVMs). Identification under additive noise case is also considered and it is shown that identification rates reduces dramatically in case of additive noise.

show abstract

mentioning

confidence: 99%

Source cell-phone recognition from recorded speech using non-speech segments

Hanilçi

Kinnunen

2014

Digital Signal Processing

Self Cite

View full text Add to dashboard Cite

show abstract

“…It is observed from the detailed review of the existing Cepstral coefficients that the two parameters such as the conversion of frequency from linear scale to perceptual scale and the selection of cut-off frequencies are the determining factor for the performance of the cepstral analysis. This issue is addressed in the literature [13] , [14] , [15] , [16] , [17] , [18] , [19] . This problem has been further investigated in Section 5.2 .…”

Section: Related Workmentioning

confidence: 99%

“…The use of these features achieves faster convergence of the ANN model and improvement in recognition accuracy at different SNR levels. The source recognition of Cell-Phone is carried out using the optimization of different cepstral coefficients such as Mel, linear, and Bark frequency [13] . The use of minimum and maximum frequencies of MFCC and cepstral variance normalization has enhanced the identification rate to 96.85%.…”

Section: Introductionmentioning

confidence: 99%

Detection of COVID-19 from speech signal using bio-inspired based cepstral features

Dash¹,

Mishra²,

Panda³

et al. 2021

Pattern Recognition

View full text Add to dashboard Cite

“…A best accuracy of 96.42% was obtained on their close set. In [3], an SVM classifier was used to compare the performance of various acoustic features, including Mel-frequency, linear frequency, bark frequency cepstral coefficients (MFCCs, LFCCs and BFCCs, respectively) as well as the linear prediction cepstral coefficients (LPCCs). Their study showed that MFCCs achieved the best performance for source identification.…”

Section: Introductionmentioning

confidence: 99%

Anti-Forensics of Audio Source Identification Using Generative Adversarial Network

Yan

Dong

et al. 2019

IEEE Access

View full text Add to dashboard Cite

Digital audio recording is the main evidence used in the field of judicial forensics. Recently, a number of digital audio forensic techniques have been developed and the audio source identification (ASI) is one of the most active research topics. Most of existing ASI works mainly focus on improving the performance of detection accuracy and robustness. Little consideration has been given to ASI anti-forensics, which aims at attacking the forensic techniques. To expose the weaknesses of these source identification methods, we propose an anti-forensic framework based on generative adversarial network (GAN) to falsify the source information of an audio by adding specific disturbance. The experimental results show that the falsified audio can deceive the forensic methods effectively, and can even control their conclusions. Three state-of-art ASI methods have been evaluated as the attacking targets. For the confusing attack, the proposed method can significantly reduce their detection accuracies from about 97% to less than 5%. For the misleading attack, a misleading rate about 81.32% has been achieved while ensuring the perceptual quality of the anti-forensic audio.

show abstract

Optimizing acoustic features for source cell-phone recognition using speech signals

Cited by 17 publications

References 31 publications

Source cell-phone recognition from recorded speech using non-speech segments

Source cell-phone recognition from recorded speech using non-speech segments

Detection of COVID-19 from speech signal using bio-inspired based cepstral features

Anti-Forensics of Audio Source Identification Using Generative Adversarial Network

Contact Info

Product

Resources

About