Speaker Identity Verification Research Articles

A Comprehensive Review of Forensic Phonetics Techniques

Forensics refers to scientific investigations of criminal activities. In the digital forensic domain, phonetics is gaining attention from researchers. In the past, it was not as highlighted as images or videos. Considering the importance of voice as a powerful means of communication in the digital world, it has become an interesting topic for research. For instance, speaker identity verification is crucial in banks, multimedia, and biometrics. However, voice can be easily tampered with or disguised, creating social panic or affecting the speaker’s identity. This paper provides a comprehensive overview of voice modification in speaker identification, making a valuable contribution to criminal investigations. It covers various topics, including voice disguise methods, analysis techniques used to detect disguises and forensic approaches for solving criminal cases in phonetics. The paper aims to understand these interrelated fields better and identify potential avenues for future research and development. It achieves this by providing a detailed review of the methodology used to identify disguised voices in forensic investigations and presenting a tabular overview of the developments in this area over the years.

Comparison of the Ability of Neural Network Model and Humans to Detect a Cloned Voice

The vulnerability of the speaker identity verification system to attacks using voice cloning was examined. The research project assumed creating a model for verifying the speaker’s identity based on voice biometrics and then testing its resistance to potential attacks using voice cloning. The Deep Speaker Neural Speaker Embedding System was trained, and the Real-Time Voice Cloning system was employed based on the SV2TTS, Tacotron, WaveRNN, and GE2E neural networks. The results of attacks using voice cloning were analyzed and discussed in the context of a subjective assessment of cloned voice fidelity. Subjective test results and attempts to authenticate speakers proved that the tested biometric identity verification system might resist voice cloning attacks even if humans cannot distinguish cloned samples from original ones.

A Lempel-Ziv like approach for signal classification

In this paper, the seminal method proposed by Abraham Lempel and Jacob Ziv, aimed at the complexity analysis of sequences of symbols, was modified to compare similarities between two sequences. This modification allowed the creation of a new criterion which can replace likelihood in some pattern recognition applications. Moreover, to allow for analysis and comparison of multivariate continuously valued patterns, we also present a simple adaptation of the Lempel-Ziv's method to time-sampled signals. To illustrate the usefulness of these proposed tools, two sets of experimental results are presented, namely: one on speaker identity verification (biometrics) and another on healthcare signal detection. Both experiments yield promising performances. Moreover, as compared to a conventional pattern recognition method, the new approach provided better performances in terms of Equal Error Ratio in speaker verification experiments.

On the use of quality measures in face and speaker identity verification based on video and audio streams

This study addresses the advantage of adding quality information of the biometric signals into a multimedia-based (video and audio) identity verification system. The quality information of the biometric signals can be used in several ways and stages in the biometric system. In this study, the authors introduce quality-based decisions in two stages: score normalisation and frame selection. Quality-based score normalisation helps to handle quality dependent drifts in the scores distributions. We derive a necessary and sufficient condition for reducing error when introducing quality-based score normalisation and present a score normalising technique. Additionally, the number of frontal faces and speech vectors extracted from the video and audio streams allows quality-based selection of frames, both in training and test, to preserve quality in the statistical representation of the signals. For these two stages we defined some quality measures for speaker and frontal face signals and run experiments to show the reliability of the proposed techniques over the BANCA database.

Robust face-voice based speaker identity verification using multilevel fusion

In this paper, we propose a robust multilevel fusion strategy involving cascaded multimodal fusion of audio–lip–face motion, correlation and depth features for biometric person authentication. The proposed approach combines the information from different audio–video based modules, namely: audio–lip motion module, audio–lip correlation module, 2D + 3D motion-depth fusion module, and performs a hybrid cascaded fusion in an automatic, unsupervised and adaptive manner, by adapting to the local performance of each module. This is done by taking the output-score based reliability estimates (confidence measures) of each of the module into account. The module weightings are determined automatically such that the reliability measure of the combined scores is maximised. To test the robustness of the proposed approach, the audio and visual speech (mouth) modalities are degraded to emulate various levels of train/test mismatch; employing additive white Gaussian noise for the audio and JPEG compression for the video signals. The results show improved fusion performance for a range of tested levels of audio and video degradation, compared to the individual module performances. Experiments on a 3D stereovision database AVOZES show that, at severe levels of audio and video mismatch, the audio, mouth, 3D face, and tri-module (audio–lip motion, correlation and depth) fusion EERs were 42.9%, 32%, 15%, and 7.3%, respectively, for biometric person authentication task.

Speaker identity verification of spectrally ambiguous and distinguishable speech

Feustel et al. [J. Acoust. Soc. Am. Suppl. 1 83, S55 (1988)] evaluated performance of humans in a speaker identity verification (SIV) task with the intent of improving the performance of a LPC‐cepstral based verification algorithm. This study uses the general methodology and signal processing techniques developed by Feustel et al. to further investigate the nature of potential invariant acoustic information (within a speaker) used by humans to verify a speaker's identity. Subjects were presented natural speech or modified natural speech stimuli drawn from a corpus of spectrally distinguishable and spectrally ambiguous utterances. The results suggest that humans can use information from both the glottal (source) waveform and spectral envelope. Overall, the glottal waveform provides more reliable information. The results will be discussed in this context.

Speaker identity feature combination with a neural net

This study compares the performance of a neural net approach with conventional linear methods as applied to the problem of feature combination in the domain of speaker identity verification (SIV). The experiment endeavors to combine features consisting of LPC-cepstral coefficient differences and pitch differences for isolated words in a template-matching scenario. The signal features are analyzed for 30-ms frames every 10 ms. The pitch estimate is based on the cepstrum of the LPC residual. Previous work [G. Velius, ICASSP 88, 583–586 (1988)] showed that the Fisher linear discriminant (FLD) was better at feature weighting (for cepstral coefficients only) than several other common linear methods. Results show that, when feature combination is done by the neural net, the SIV task is performed significantly better than when the feature combination (i.e., weighting) is done by the FLD. The neural network architecture used in this experiment was in no way “optimized” for the specific task at hand. An additional finding is that the pitch feature used here, in conjuction with the cepstral coefficients, contributes significantly to the SIV task; that is, the error rate is reduced by 13%.

Human and machine performance on speaker identity verification

Two experiments were conducted to identify acoustic features for speaker identity verification (SIV) that are used by humans and not by cepstral-based algorithms. Although these algorithms generally out-perform human listeners for randomly selected comparisons between single-word utterances, this approach was to analyze human performance on comparisons that could not be effectively discriminated by machine. Experiment 1 showed that humans could perform at high levels of accuracy on these comparisons suggesting that either information exists that is not captured by the algorithms, or that the information is coded by the algorithms but is not used effectively. The second experiment consisted of three stimulus conditions for SIV; digitized speech signals, noise-excited resynthesized LPC signals, and error prediction signals from the LPC. Results indicated high levels of performance in the natural and error prediction signal conditions and performance near chance in the noise excited condition, thus suggesting that the error signal provides valuable information that allows humans to distinguish between speakers. It may be possible to improve verification algorithms by adapting current models to more accurately utilize information used by human listeners.

Long‐term changes in voice characteristics: Implications for speaker identity verification (SIV)

In a typical speaker identity verification (SIV) scenario, a talker first asserts an identity claim via some means other than speech. A reference sample of the claimed individual's speech is then loaded into a pattern matcher and compared against a new sample of the same word or phrase solicited on‐line. Based on the similarity of the two utterances, and a predetermined criterion, a decision is made as to the veracity of the claimed identity. One difficult problem for SIV applications is that the spectral characteristics of a speaker's voice change over time in ways that are poorly understood. Utterances were collected from three speakers at regular intervals over a period of 23 weeks, and were analyzed first with respect to the magnitude of spectral dissimilarity over varying time intervals, and, second, with respect to the effect of these differences on error rates for an SIV decision task. The data show an increase in mean utterance dissimilarity of about 8% over a period of 2 weeks, followed by a much more gradual increase over 6 months. While the absolute change in mean dissimilarity was small, it resulted in a doubling of the error rate for the SIV decision task from about 5% to 10%.In a typical speaker identity verification (SIV) scenario, a talker first asserts an identity claim via some means other than speech. A reference sample of the claimed individual's speech is then loaded into a pattern matcher and compared against a new sample of the same word or phrase solicited on‐line. Based on the similarity of the two utterances, and a predetermined criterion, a decision is made as to the veracity of the claimed identity. One difficult problem for SIV applications is that the spectral characteristics of a speaker's voice change over time in ways that are poorly understood. Utterances were collected from three speakers at regular intervals over a period of 23 weeks, and were analyzed first with respect to the magnitude of spectral dissimilarity over varying time intervals, and, second, with respect to the effect of these differences on error rates for an SIV decision task. The data show an increase in mean utterance dissimilarity of about 8% over a period of 2 weeks, followed by a much...