Academia.eduAcademia.edu

Outline

Title
Abstract
Introduction
Mobiphone Database
Experimental Evaluation
Conclusions
References
All Topics
Computer Science

Mobile phone identification using recorded speech signals

Profile image of Constantine KotropoulosConstantine Kotropoulos

2014, 2014 19th International Conference on Digital Signal Processing

https://doi.org/10.1109/ICDSP.2014.6900732
visibility

…

description

6 pages

link

1 file

Abstract

In this paper, we elaborate on mobile phone identification from recorded speech signals. The goal is to extract intrinsic traces related to the mobile phone used to record a speech signal. Mel frequency cepstral coefficients (MFCCs) are extracted from any recorded speech signal at a frame level. The sequences of the MFCC vectors extracted from each recording device train a Gaussian Mixture Model with diagonal covariance matrices. A Gaussian supervector is derived by concatenating the mean vectors and the main diagonals of the covariance matrices that is used as a template for each device. Experiments were conducted on a database of 21 mobile phones of various models from 7 different brands. The aforementioned database, that is called MOBIPHONE, was collected by recording 10 utterances, uttered by 12 male speakers and another 12 female speakers, randomly chosen from the TIMIT database. Three commonly used classifiers were employed, such as Support Vector Machines with different kernels, a Radial Basis Functions neural network, and a Multi-Layer Perceptron. The best identification accuracy (97.6%) was obtained by the Radial Basis Functions neural network.

Related papers

Improving of Speaker Identification from Mobile Telephone Calls
Tomasz Marciniak

Communications in Computer and Information Science, 2014

The paper examines issues related to proper selection of models used for quick speaker recognition based on short recordings of mobile telephone conversations. A knowledge of the encoder type used during the transmission of speech allows to apply an appropriate model that takes specific characteristics of the encoder into account: full rate (FR), half rate (HR), enhanced full rate (EFR) and adaptive multi-rate (AMR). We analyse both proper model selection and automatic silence removal. Analysis of time of processing is also a part of this study.

View PDFchevron_right
Individual Identification Through Voice Using Mel-Frequency Cepstrum Coefficient (MFCC) and Hidden Markov Models (HMM) Method
sofia saidah

Journal of Measurements, Electronics, Communications, and Systems, 2020

Voice is one of the parameters in the identification process of a person. Through the voice, information will be obtained such as gender, age, and even the identity of the speaker. Speaker recognition is a method to narrow down crimes and frauds committed by voice. So that it will minimize the occurrence of faking one's identity. The Method of Mel Frequency Cepstrum Coefficient (MFCC) can be used in the speech recognition system. The process of feature extraction of speech signal using MFCC will produce acoustic speech signal. The classification, Hidden Markov Models (HMM) is used to match unidentified speaker’s voice with the voices in database. In this research, the system is used to verify the speaker, namely 15 text dependent in Indonesian. On testing the speaker with the same as database, the highest accuracy is 99,16%.

View PDFchevron_right
Speaker Identification using Mel Frequency Cepstral Coefficient and BPNN
DEBANANDA PADHI

Speech processing is emerged as one of the important application area of digital signal processing. Various fields for research in speech processing are speech recognition, speaker recognition, speech synthesis, speech coding etc. The objective of automatic speaker recognition is to extract, characterize and recognize the information about speaker identity. Feature extraction is the first step for speaker recognition. Many algorithms are suggested/developed by the researchers for feature extraction. In this work, the Mel Frequency Cepstrum Coefficient (MFCC) feature has been used for designing a text dependent speaker identification system. BPNN is used for identification of speaker after training the feature set from MFCC. Some modifications to the existing technique of MFCC for feature extraction are also suggested to improve the speaker recognition efficiency. Information from speech recognition can be used in various ways in state-of-the-art speaker recognition systems. This includes the obvious use of recognized words to enable the use of text-dependent speaker modeling techniques when the words spoken are not given. Furthermore, it has been shown that the choice of words and phones itself can be a useful indicator of speaker identity. Also, recognizer output enables higher-level features, in particular those related to prosodic properties of speech.

View PDFchevron_right
The structural tuning of the convolutional neural network for speaker identification in mel frequency cepstrum coefficients space. Anastasiia D. Matychenko, Marina V. Polyakova (Herald of Advanced Information Technology, Vol. 6 No. 2)
Anastasiia D. Matychenko, Herald of Advanced Information Technology

The structural tuning of the convolutional neural network for speaker identification in mel frequency cepstrum coefficients space, 2023

As a result of the literature analysis, the main methods for speaker identification from speech signals were defined. These are statistical methods based on Gaussian mixture model and a universal background model, as well as neural network methods, in particular, using convolutional or Siamese neural networks. The main characteristics of these methods are the recognition performance, a number of parameters, and the training time. High recognition performance is achieved by using convolutional neural networks, but a number of parameters of these networks are much higher than for statistical methods, although lower than for Siamese neural networks. A large number of parameters require a large training set, which is not always available for the researcher. In addition, despite the effectiveness of convolutional neural networks, model size and inference efficiency remain important for devices with a limited source of computing power, such as peripheral or mobile devices. Therefore, the aspects of tuning of the structure of existing convolutional neural networks are relevant for research. In this work, we have performed a structural tuning of an existing convolutional neural network based on the VGGNet architecture for speaker identification in the space of mel frequency cepstrum coefficients. The aim of the work was to reduce the number of neural network parameters and, as a result, to reduce the network training time, provided that the recognition performance is sufficient (the correct recognition is above 95 %). The neural network proposed as a result of structural tuning has fewer layers than the architecture of the basic neural network. Instead of the ReLU activation function, the related Leaky ReLU function with a parameter of 0.1 was used. The number of filters and the size of kernels in convolutional layers are changed. The size of kernels for the max pooling layer has been increased. It is proposed to use the averaging of the results of each convolution to input a two-dimensional convolution results to a fully connected layer with the Softmax activation function. The performed experiment showed that the number of parameters of the proposed neural network is 29 % less than the number of parameters of the basic neural network, provided that the speaker recognition performance is almost the same. In addition, the training time of the proposed and basic neural networks was evaluated on five datasets of audio recordings corresponding to different numbers of speakers. The training time of the proposed network was reduced by 10-39 % compared to the basic neural network. The results of the research show the advisability of the structural tuning of the convolutional neural network for devices with a limited source of computing, namely, peripheral or mobile devices.

View PDFchevron_right
Mel-Frequency Cepstrum Coeffficients as Higher Order Statistics Representation to Characterize Speech Signal for Speaker Identification System in Noisy Environment Using Hidden Markov Model
Benyamin Kusumoputro
View PDFchevron_right
Applying MFCC-based automatic speaker recognition to GSM and forensic data
Tomi Kinnunen

Speaker Profiler computer program for automatic speaker recognition has been developed in a research project funded by the Finnish Technology Agency. A vector quantization (VQ) matching approach is used, where dissimilarity of an unknown speech sample is computed for codebooks created using the K-means algorithm. This study tests the recognition reliability with two databases constructed from Finnish band-limited GSM speech and authentic crime case speech. Material for the first test is recorded with a GSM phone and a laptop computer. Spontaneous speech vs. reading was tested. The program should pick the right person from the database based on independent non-verbatim speech samples. There were 47.5 % out of 107 samples ranked first correctly. Some very poor quality speech files were used in training and the mother tongue for some speakers was not Finnish. If these samples were not considered, the result was better. The second part of this study consists of real crime investigation ...

View PDFchevron_right
Speaker, accent, and language identification using multilingual phone strings
Kornel Laskowski

2002

View PDFchevron_right
Automatic Speaker Recognition for Series 60 Mobile Devices
Ville Hautamäki

Mobile phone environment and other devices that require low power consumption, restrict the computations of DSP processors to fixed point arithmetic only. This is a common practise based on cost and battery power savings. In this paper, we first discuss the specific software architecture requirements set up by the Symbian operating system and the Series 60 platform. We analyse mel frequency cepstral coefficient based classification, and show techniques to avoid information loss, when a floating-point algorithm is replaced by a corresponding algorithm that uses fixed-point arithmetic. We analyse the preservation of discrimination information, which is the key motivation in all classification applications. We also give insight to the relation between information preserving and operator presentation accuracy. The results are exemplified with tests made on algorithms that are identical except for the different arithmetics used.

View PDFchevron_right
Use of Mel Frequency Cepstral Coefficients for the Implementation of a Speaker Recognition System
Bhoomika Purohit

2019

The paper proposes a Speaker Recognition system which does the task of validating a user’s claimed identity using characteristics extracted from their voices. It is one of the most useful and popular biometric recognition techniques in the world especially related to areas in which security is a major concern. A direct analysis and synthesizing of the complex voice signal is due to too much information contained in the signal. Therefore, the digital signal processes, Feature Extraction and Feature Matching were introduced to represent the voice signal. MelFrequency Cepstral Coefficients (MFCC) were extracted from the speech signal which were used to represent each speaker and recognition was carried out using weighted Euclidean distance. MATLABR2017b platform was used to implement feature extraction process. Index Terms – Co Feature matching, Feature Extraction, MFCC, Euclidean distance.

View PDFchevron_right
A phone-based approach to non-linguistic speech feature identification
Lori Lamel

Computer Speech & Language, 1995

In this paper we present a general approach to identifying non-linguistic speech features from the recorded signal using phone-based acoustic likelihoods. The basic idea is to process the unknown speech signal by feature-specific phone model sets in parallel, and to hypothesize the feature value associated with the model set having the highest likelihood. This technique is shown to be effective for text-independent gender, speaker, and language identification. Text-independent speaker identification accuracies of 98.8% on TIMIT (168 speakers) and 99.2% on BREF (65 speakers), were obtained with one utterance per speaker, and 100% with 2 utterances for both corpora. Experiments in which speaker-specific models were estimated without using of the phonetic transcriptions for the TIMIT speakers had the same identification accuracies obtained with the use of the transcriptions. French/English language identification is better than 99% with 2s of read, laboratory speech. On spontaneous telephone speech from the OGI corpus, the language can be identified as French or English with 82% accuracy with 10s of speech. 10 language identification using the OGI corpus is 59.7% with 10s of signal.

View PDFchevron_right

References (34)

  1. D. Garcia-Romero and C. Y. Espy-Wilson, "Automatic acquisition device identification from speech recordings," in Proc. 2010 IEEE Int. Conf. Acoustics, Speech, and Signal Processing, Dallas, TX, USA, 2010, pp. 1806-1809.
  2. C. Hanilci, F. Ertas, T. Ertas, and O. Eskidere, "Recognition of brand and models of cell-phones from recorded speech signals," IEEE Trans. Information Forensics and Security, vol. 7, no. 2, pp. 625-634, 2012.
  3. R. Maher, "Audio forensic examination," IEEE Signal Processing Mag- azine, vol. 26, no. 2, pp. 84-94, 2009.
  4. Proceedings of the 19th International Conference on Digital Signal Processing 20-23 August 2014 978-1-4799-4612-9/14/$31.00 © 2014 IEEE 590 DSP 2014
  5. R. Yang, Z. Qu, and J. Huang, "Detecting digital audio forgeries by checking frame offsets," in Proc. 10th ACM Multimedia and Security Workshop, New York, NY, USA, 2008, pp. 21-26.
  6. D. Luo, W. Luo, R. Yang, and J. Huang, "Compression history identifi- cation for digital audio signal," in Proc. 2012 IEEE Int. Conf. Acoustics, Speech, and Signal Processing, Kyoto, Japan, 2012, pp. 1733-1736.
  7. J. Zhou, D. Garcia-Romero, and C. Y. Espy-Wilson, "Automatic speech codec identification with applications to tampering detection of recordings," in Proc. 12th INTERSPEECH, Florence, Italy, 2011, pp. 2533-2536.
  8. F. Jenner and A. Kwasinski, "Highly accurate non-intrusive speech forensics for codec identifications from observed decoded signals," in Proc. 2012 IEEE Int. Conf. Acoustics, Speech, and Signal Processing, Kyoto, Japan, 2012, pp. 1737-1740.
  9. D. Sharma, P. A. Naylor, N. D. Gaubitch, and M. Brookes, "Non intrusive codec identification algorithm," in Proc. 2012 IEEE Int. Conf. Acoustics, Speech, and Signal Processing, Kyoto, Japan, 2012, pp. 4477- 4480.
  10. A. Oermann, A. Lang, and J. Dittmann, "Verifier-tuple for audio-forensic to determine speaker environment," in Proc. 7th ACM Multimedia and Security Workshop, New York, NY, USA, 2005, pp. 57-62.
  11. C. Kraetzer, A. Oermann, J. Dittmann, and A. Lang, "Digital audio forensics: a first practical evaluation on microphone and environment classification," in Proc. 9th ACM Multimedia and Security Workshop, Dallas, TX, USA, 2007, pp. 63-74.
  12. C. Kraetzer, M. Schott, and J. Dittmann, "Unweighted fusion in mi- crophone forensics using a decision tree and linear logistic regression models," in Proc. 11th ACM Multimedia and Security Workshop, Prince- ton, NJ, USA, 2009, pp. 49-56.
  13. H. Malik and H. Farid, "Audio forensics from acoustic reverberation," in Proc. 2010 IEEE Int. Conf. Acoustics Speech and Signal Processing, Dallas, TX, USA, 2010, pp. 1710-1713.
  14. C. C. Huang and J. Epps, "A study of automatic phonetic segmentation for forensic voice comparison," in Proc. 2012 IEEE Int. Conf. Acoustics, Speech, and Signal Processing, Kyoto, Japan, 2012, pp. 1853-1856.
  15. H. Zhao and H. Malik, "Acoustic recording location identification using acoustic environment signature," IEEE Trans. Information Forensics and Security, vol. 8, no. 11, pp. 1746-1759, 2013.
  16. H. Malik, "Acoustic environment identification and its applications to audio forensics," IEEE Trans. Information Forensics and Security, vol. 8, no. 11, pp. 1827-1837, 2013.
  17. A. J. Cooper, "Further considerations for the analysis of ENF data for forensic audio and video applications," The Int. J. of Speech, Language, and the Law, vol. 18, no. 1, pp. 99-120, 2011.
  18. O. Ojowu, Jr., J. Karlsson, J. Li, and Y. Liu, "ENF extraction from digital recordings using adaptive techniques and frequency tracking," IEEE Trans. Information Forensics and Security, vol. 7, no. 4, pp. 1330- 1338, 2012.
  19. V. A. Balasubramaniyan, P. Amit, M. Ahamad, M. Hunter, and P. Tranyo, "PinDr0p: Using single-ended audio features to determine call prove- nances," in Proc. 17th ACM Conf. Computer Communications, Chicago, IL, 2010, pp. 109-120.
  20. S. Kishore and Y. B., "Identification of handset type using autoasso- ciative neural networks," in Proc. 4th Int. Conf. Advances in Pattern Recognition and Digital Techniques, 1999, pp. 353-356.
  21. M.-W. Mak and S.-Y. Kung, "Combining stochastic feature transforma- tion and handset identification for telephone-based speaker verification," in Proc. 2002 IEEE Int. Conf. Acoustics, Speech, and Signal Processing, vol. I, Orlando, FL, 2002, pp. 701-704.
  22. S. B. Davies and P. Mermelstein, "Comparison of parametric repre- sentations for monosyllabic word recognition in continuously spoken sentences," IEEE Trans. ASSP, vol. 28, no. 4, pp. 357-366, 1980.
  23. D. Reynolds, "HTIMIT and LLHDB: speech corpora for the study of handset transducer effects," in Proc. 1997 IEEE Int. Conf. Acoustics, Speech, and Signal Processing, vol. 2, Munich, Germany, 1997, pp. 1535-1538.
  24. Y. Panagakis and C. Kotropoulos, "Automatic telephone handset identi- fication by sparse representation of random spectral features," in Proc. 14th ACM Multimedia and Security Workshop, Coventry, U.K., 2012, pp. 91-95.
  25. --, "Telephone handset identification by feature selection and sparse representations," in Proc. 2012 IEEE Int. Workshop Information Foren- sics and Security, Tenerife, Spain, 2012, pp. 73-78.
  26. C. Kotropoulos, "Source phone identification using sketches of features," IET Biometrics, 2014.
  27. D. A. Reynolds, T. F. Quatieri, and R. Dunn, "Speaker verification using adapted gaussian mixture models," Digital Signal Processing, vol. 10, no. 1-3, pp. 19-41, 2000.
  28. J. Garofolo, "Getting started with the DARPA TIMIT cd-rom: An acoustic phonetic continuous speech database," National Inst. Standards and Technology (NIST), Tech. Rep., 1988.
  29. J. R. Deller, Jr., J. H. L. Hansen, and J. G. Proakis, Discrete-Time Processing of Speech Signals. New York, NY, USA: Wiley-Interscience- IEEE, 2000.
  30. A. Dempster, N. Laird, and D. Rubin, "Maximum likelihood from incomplete data via the em algorithm (with discussion)," Journal Royal Statistical Society, Series B, vol. 39, pp. 1-38, 1977.
  31. V. Vapnik, Statistical Learning Theory. New York, NY, USA: J. Wiley & Sons, 1998.
  32. C.-C. Chang and C.-J. Lin, "LIBSVM: A library for support vector machines," ACM Transactions Intelligent System Technologies, vol. 2, no. 3, pp. 1-27, 2011.
  33. S. Haykin, Neural Networks and Learning Machines, 3/e. Upper Saddle River, N.J., USA: Prentice Hall, 2008.
  34. I. Guyon, J. Makhoul, R. Schwartz, and V. Vapnik, "What size test set gives good error rate estimates?" IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 20, no. 1, pp. 52-64, 1998.

Related papers

Blind source mobile device identification based on recorded call
Mohd Idris, Mehdi Jahanirad

Engineering Applications of Artificial Intelligence, 2014

Mel-frequency cepstrum coefficients (MFCCs) extracted from speech recordings has been proven to be the most effective feature set to capture the frequency spectra produced by a recording device. This paper claims that audio evidence such as a recorded call contains intrinsic artifacts at both transmitting and receiving ends. These artifacts allow recognition of the source mobile device on the other end through recording the call. However, MFCC features are contextualized by the speech contents, speaker's characteristics and environments. Thus, a device-based technique needs to consider the identification of source transmission devices and improve the robustness of MFCCs. This paper aims to investigate the use of entropy of Mel-cepstrum coefficients to extract intrinsic mobile device features from near-silent segments, where it remains robust to the characteristics of different speakers. The proposed features are compared with five different combinations of statistical moments of MFCCs, including the mean, standard deviation, variance, skewness, and kurtosis of MFCCs. All feature sets are analyzed by using five supervised learning techniques, namely, support vector machine, naïve Bayesian, neural network, linear logistic regression, and rotation forest classifier, as well as two unsupervised learning techniques known as probabilistic-based and nearest-neighbor-based algorithms. The experimental results show that the best performance was achieved with entropy-MFCC features that use the naïve Bayesian classifier, which resulted in an average accuracy of 99.99% among 21 mobile devices.

View PDFchevron_right
Source cell-phone recognition from recorded speech using non-speech segments
Tomi Kinnunen

Digital Signal Processing, 2014

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.

View PDFchevron_right
Phone Recognition on TIMIT Database Using Mel-Frequency Cepstral Coefficients and Support Vector Machine Classifier
vahid hosseinzadeh

2015

In the present article the speech recognition is performed on 61 English phones extracted from TIMIT database. For this reason the speech signal of each phone is divided into 20 millisecond frames and for each frame, the MFCC coefficients, their first and second derivatives, energy of the frame, as well as its first derivative were extracted as 41 features of the vector frame and by the use of the support vector machines, the frame classification into English phonetics is performed. In order to improve the results the sinusoidal Lifter and Independent Component Analysis (ICA) are applied on feature vectors. According to the results of the study, there have been 4.93% error rate for the test data which presents a significant improvement compared with the recent researches in this area and demonstrate the high performance of the features and algorithms used in the article.

View PDFchevron_right
Identification from Recorded Calls
Mehdi Jahanirad

2014

Keywords: This study investigates the use of blind source computer device identification for forensic investigation of the recorded VoIP call. It was found that a combination of mel-frequency cepstrum coefficients (MFCCs) and entropy as an intrinsic audio feature captures the specific frequency response due to the tolerance in the nominal values of the electronic components associated to individual computer device. By applying the supervised learning techniques such as naïve Bayesian, linear logistic regression, neural networks (NN), support vector machines (SVM) and sequential minimal optimization (SMO) classifier to the Entropy-MFCC features, state-of-the-art identification accuracy of above 99.8% has been achieved on a set of 5 iMacs and 5 desktop PCs from the same model.

View PDFchevron_right
Telephone Handset Identification by Collaborative Representations
Constantine Kotropoulos

Recorded speech signals convey information not only for the speakers' identity and the spoken language, but also for the acquisition devices used for their recording. Therefore, it is reasonable to perform acquisition device identification by analyzing the recorded speech signal. To this end, recording-level spectral, cepstral, and fusion of spectral and cepstral features are employed as suitable representations for device identification. The feature vectors extracted from the training speech recordings are used to form overcomplete dictionaries for the devices. Each test feature vector is represented as a linear combination of all the dictionary columns (i.e., atoms). Since the dimensionality of the feature vectors is much smaller than the number of training speech recordings, there are infinitely many representations of each test feature vector with respect to the dictionary. These representations are referred to as collaborative representations in the sense that all the dictionary atoms collaboratively represent any test feature vector. By imposing the representation to be either sparse (i.e., to admit the minimum 1 norm) or to have the minimum 2 norm, unique collaborative representations are obtained. The classification is performed by assigning each test feature vector the device identity of the dictionary atoms yielding the minimum reconstruction error. This classification method is referred to as the sparse representation-based classifier (SRC) if the sparse collaborative representation is employed and as the least squares collaborative representation-based classifier (LSCRC) in the case of the minimum 2 norm regularized collaborative representation is used for

View PDFchevron_right

Related topics

  • Computer Science

    • Cited by

    Digital Recording System Identification Based on Blind Deconvolution
    Dimitrios Pados

    2019

    In this work, we develop a theoretical framework for reliable digital recording system identification from digital audio files alone, for forensic purposes. A digital recording system consists of a microphone and a digital sound processing card. We view the cascade as a system of unknown transfer function. We expect the same manufacturer and model microphone-sound card combinations to have very similar/near identical transfer functions, bar any unique manufacturing defect. Input voice (or other) signals are modeled as non-stationary processes. The technical problem under consideration becomes blind deconvolution with non-stationary inputs, as it manifests itself in the specific application of digital audio recording equipment classification. Experimental results demonstrate over 99.2% accuracy in identification of the recording devices.

    View PDFchevron_right
    VoIPLoc
    ryan shah

    Proceedings of the 14th ACM Conference on Security and Privacy in Wireless and Mobile Networks, 2021

    We propose VoIPLoc, a novel location fingerprinting technique and apply it to the VoIP call provenance problem. It exploits echo-location information embedded within VoIP audio to support fine-grained location inference. We found consistent statistical features induced by the echo-reflection characteristics of the location into recorded speech. These features are discernible within traces received at the VoIP destination, enabling location inference. We evaluated VoIPLoc by developing a dataset of audio traces received through VoIP channels over the Tor network. We show that recording locations can be fingerprinted and detected remotely with a low false-positive rate, even when a majority of the audio samples are unlabelled. Finally, we note that the technique is fully passive and thus undetectable, unlike prior art. VoIPLoc is robust to the impact of environmental noise and background sounds, as well as the impact of compressive codecs and network jitter. The technique is also highly scalable and offers several degrees of freedom terms of the fingerprintable space. CCS CONCEPTS • Networks → Web protocol security; Network security; • Security and privacy → Pseudonymity, anonymity and untraceability.

    View PDFchevron_right
    580 California St., Suite 400
    San Francisco, CA, 94104
    © 2025 Academia. All rights reserved