Voiceprint Identification Research Articles

Research on blind source separation of operation sounds of metro power transformer through an Adaptive Threshold REPET algorithm

PurposeThe safe operation of the metro power transformer directly relates to the safety and efficiency of the entire metro system. Through voiceprint technology, the sounds emitted by the transformer can be monitored in real-time, thereby achieving real-time monitoring of the transformer’s operational status. However, the environment surrounding power transformers is filled with various interfering sounds that intertwine with both the normal operational voiceprints and faulty voiceprints of the transformer, severely impacting the accuracy and reliability of voiceprint identification. Therefore, effective preprocessing steps are required to identify and separate the sound signals of transformer operation, which is a prerequisite for subsequent analysis.Design/methodology/approachThis paper proposes an Adaptive Threshold Repeating Pattern Extraction Technique (REPET) algorithm to separate and denoise the transformer operation sound signals. By analyzing the Short-Time Fourier Transform (STFT) amplitude spectrum, the algorithm identifies and utilizes the repeating periodic structures within the signal to automatically adjust the threshold, effectively distinguishing and extracting stable background signals from transient foreground events. The REPET algorithm first calculates the autocorrelation matrix of the signal to determine the repeating period, then constructs a repeating segment model. Through comparison with the amplitude spectrum of the original signal, repeating patterns are extracted and a soft time-frequency mask is generated.FindingsAfter adaptive thresholding processing, the target signal is separated. Experiments conducted on mixed sounds to separate background sounds from foreground sounds using this algorithm and comparing the results with those obtained using the FastICA algorithm demonstrate that the Adaptive Threshold REPET method achieves good separation effects.Originality/valueA REPET method with adaptive threshold is proposed, which adopts the dynamic threshold adjustment mechanism, adaptively calculates the threshold for blind source separation and improves the adaptability and robustness of the algorithm to the statistical characteristics of the signal. It also lays the foundation for transformer fault detection based on acoustic fingerprinting.

Text-independent voiceprint recognition via compact embedding of dilated deep convolutional neural networks

In order to process speech, most state-of-the-art experimental methods employ convolutional neural networks (CNNs), which operate on a continuous, 1-dimensional (1-D) time stream. In an audio signal, the mel-spectrogram facilitates the representation of attributes of the utterances' in the frequency domain (which corresponds to the speech spectrum). Moreover, for a time-series speaker signal, CNNs are superior to machine or transfer learning models in capturing characteristics from long-form talks. This paper introduces a jump-connected 1-D CNN that employs a combined loss function for speaker recognition. The suggested model uses a 1-D convolutional layer combined with jump connections to extract speaker-specific characteristics; this reduces time-based and frequency-based variability for faster computing. A combined softmax loss, stable L2-norm, and smooth L1-norm loss function guide the proposed compact convolutional neural networks (CCNN) to identify the correct spokesman with improved efficacy. We evaluated the proposed framework using various standard and real-time audio datasets. The experimental findings demonstrate that the proposed CCNN outperforms existing approaches by reducing the equal error rate by 9.02 %. Also, our recommended voiceprint identification model achieves an impressive average speaker recognition rate of 98.76 %. Simultaneously, the reliability of the 1-D CCNN is tested under various conditions. Other fields of study, like language modelling, could employ this approach after some fine-tuning.Relevance of the work: Speaker recognition is an area of interest in which machine learning (ML) and deep learning (DL) schemes, when combined, have the potential to make history in the areas of forensic sciences, automation, and authentication. Using a modest CNN can enhance the identification and verification process by ignoring many issues such as false positives, background noise, and so on. Expanding this process would facilitate raga identification and disease treatment therapies.

An adaptive algorithm for voice quality based on big data voiceprint identification

Abstract This paper improves the speech amplitude in Bayesian speech enhancement estimation by introducing a super-Gaussian cardinality distribution probability density function in the filter's construction. The derivation is combined with the perceptual error function, the new probability density function and the perceptual error cost function to better exploit and utilize the prior statistical information of the speech. The results show that the proposed method can improve the signal-to-noise ratio up to 0.7 dB under different noises and different signal-to-noise ratios, and the processed speech has better feasibility, which provides good speech enhancement for the processing of noisy speech quality in vocal identification practice without significantly increasing the computational complexity and can be better adapted to the application.

A multi-hole resonator enhanced acoustic energy harvester for ultra-high electrical output and machine-learning-assisted intelligent voice sensing

Acoustic energy harvesters are promising platforms for noise-to-electricity conversion in distributed environmental power supply and sound sensing in intuitive voice user interface. However, their efficacy in real-world applications is frustrated by the low electric output. We here report a multi-hole acoustic resonator enhanced triboelectric nanogenerator (MHAR-TENG) that couples a perforated plate resonator and a pressure differential acoustic receiver. It generated open-circuit voltage of 347 V, short-circuit current of 95 µA, and record-high power density of 8.9 W/m2. The numerical simulation confirmed the sound pressure amplification for enhanced output performance. Besides, the high signal-to-noise ratio and precise high-frequency response of electrical signals enabled the high-quality restoration of music and human speech. Assisted by machine learning, the MHAR-TENG realized self-powered voiceprint identification and emotion recognition with a remarkably improved accuracy over 90 %. Our MHAR-TENG holds great potential for efficient acoustic-to-electrical conversion in the fields of noise recycling, biometric authentication, and human-machine interaction.

Biometric imaginaries: Formatting voice, body, identity to data

This article examines the sociotechnical imaginary within which contemporary biometric listening or VIA (voice identification and analysis) technologies are being developed. Starting from an examination of a key article on Voiceprint identification written in the 1940s, I interrogate the conceptual link between voice, body, and identity, which was central to these early attempts at technologizing voice identification. By surveying patents that delineate systems for voice identification, collection methods for voice data, and voice analysis, I find that the VIA industry is dependent on the conceptual affixion of voice to identity based on a reduction of voice that sees it as a fixed, extractable, and measurable 'sound object' located within the body. This informs the thinking of developers in the VIA industry, resulting in a reframing of the technological shortcomings of voice identification under the rubric of big data. Ultimately, this reframing rationalizes the implementation of audio surveillance systems into existing telecommunications infrastructures through which voice data is acquired on a massive scale.

Principle Analysis of Voiceprint Identification and Authentication System Based on LSTM

Recurrent neural networks, as an effective method to study the analysis and prediction of large-scale time series, have existed in various applications on time series data, but the traditional RNN neural network has the problem of gradient disappearance and the problem of poor prediction accuracy has not been solved. This paper proposes a vocal recognition system based on LSTM neural network, whose core structure can be divided into four parts: forgetting gate, input gate, cell The core structure can be divided into four parts: forgetting gate, input gate, state and output gate, which are more suitable to analyze the characteristics of different human voice patterns, and can obtain higher recognition accuracy after the process of feature extraction, data enhancement, model training and voice pattern recognition. Finally, the performance of the LSTM neural network is tested on 300 speech data, and the test results prove that the LSTM neural network can obtain high recognition rate with less iterations.

Vocal Features: From Voice Identification to Speech Recognition by Machine.

This article considers machine methods used in the collection, processing, and application of vocal recordings for speaker identification and speech recognition between 1908 and 1970. The first phonographic archives featured collections of "vocal portraits" that prompted international investigations into the essential features of human voices for individual identification. Visual records of speech later found the same applications, but as "voiceprint identification" via sound spectrography began to achieve legal and commercial success in the 1960s, the procedure attracted more widespread scientific attention, which ultimately discredited both its accuracy and its rationale. At the same time, spectrogram collections spurred a new application-speech recognition by machine. The changing status of the speech spectrogram, from a record of unique features of individual voices to a model of fundamental invariants in speech sounds, was rooted in the demands of automated processing and a corresponding shift from the sound archive to the acoustic database.

Mobile user authentication system in cloud environment

ABSTRACTIn order to reach a safe environment that can be automatically used on the Internet and to take precautions against the Internet fishing attack, the system integrates some features including one‐time password, Completely Automated Public Turing Test to tell Computers and Humans Apart, voiceprint identification of creatural features, and visual cryptography, designing a formula wherein users do not need to remember any accounts and passwords when they surf the Internet through mobile devices, and it aims at smart phones and the Cloud. The formula is able to improve the problems of rampant Internet fishing and the management of passwords. In techniques, on one hand, it uses PIN information visual passwords in cell phones to improve the security of the account; on the other hand, it uses voiceprint identification features so that the system center can ensure the user's identification with a view to improve the leak in mobile devices rather than only to check mobile devices. And then, it utilizes the voiceprint, which we use when we log in, to produce a one‐time password that is able to lower the risk of the account and passwords being attacked by Internet fishing. Through the frame of this research, it can protect our cell phones from being lost and embezzled and can prevent the account and passwords from being attacked by Internet fishing. It can also solve the problem of users forgetting accounts and passwords, and reduce the operational burden of cell phones. Besides, it is capable of preventing the Cloud servers from incurring many malicious registrations and logins, keeping them working efficiently. Copyright © 2012 John Wiley & Sons, Ltd.

Voiceprint identification and its application to sociological studies of wild Japanese monkeys (Macaca fuscata yakui)

Japanese monkeys often exchange the particular vocal sound, “coo,” especially when they feed or move as a group. It was considered that the “coo” sound had no positive social meaning, perhaps because the “coo” sound network and its function were hidden behind other behavioral observations. For identification of the vocalizer only from hearing the “coo” sound, three phonetic values, i.e., the “fundamental,” “duration,” and “formants,” plus other characteristics were used as indices of voiceprints. The results indicated that these were effective for identifying the vocalizer in two-thirds of the adults in the study troop which was composed of 12 adults and 16 immature members. The “coo” sound exchange network among the troop members (adults) was drawn on the basis of the voiceprint identification. The network showed three characteristics as follows: (1) matriarchs of the kin-groups frequently exchanged “coo” sounds with each other; (2) the other females exchanged “coo” sounds mostly within their own kin-groups; and (3) males seldom participated in the “coo” sound exchange. This suggests that “coo” sound exchange plays a central role for the matriarch of kin-groups in binding each kin-group and, ultimately, in binding all members together into an organized troop.

The first 100 years of “voiceprinting.”

“Voiceprinting,” appropriately defined, is the sound-spectro-graphic technology aspect of speech and voice individuation. A speechsound spectrogram is at once a voicesound spectrogram, since voice is the biophysical instrument of speech. Speaker individuation, known widely today as voice identification, has been with us throughout history. Technological implications of speech/voice idiosyncrasy have been sighted, and/or explored instrumentally, for at least 100 years. Here listed is a partially annotated chronology of authors and publication dates significant for this author's 30-yr commitment to the development of an instrumental speaker-individuation technology: Alexander Melville Bell (Visible Speech: The Science of … Physiological Letters, etc.): 1867; Galton (“fingerprinting”): 1892 (cf. Neumueller GREW—1684); Scripture!: 1902 (653 pp.); Liddell! (The Physical Characteristics of Speech Sound): 1924–1925–1927; Paget: 1930; Scripture (“vowel track analysis”): 1933; Lacerda: 1932–1934; Steinberg!: 1934; Liddell: 1940; Gray/Kopp! (“Voice Print Identification,” 41 pp., explicit): 1944; Potter: 1945–1946; Steinberg/French!: 1946; Koenig/Dunn/Lacy: 1946; Kopp/Green!: 1946; Potter/Kopp/Green!: 1947; Joos! (Acoustic Phonetics): 1948; Truby: 1957-1959!-1960 (“cryprinting”); Fant: 1960; Cummins/Midlo (“Dermatoglyphics”): 1961; Truby: 1962; Kersta (“Voiceprint Identification”): 1962; Garvin/Ladefoged (“Speaker Identification … Speaker Recognition”): 1963 (variously: 1950–1970: Peterson, Stevens, Flanagan, Fant, Lehiste, Truby); Tosi et al.: 1971; Smrkovski: 1974; plus the continuing, practical, day-by-day, field-laboratory investigations of Hall, Smrkovski, Chiari, Richardson, Tosi, Truby, and other Certified Voice Identification Examiners: esp., 1972–1977. And whom, after all, does the relevant “scientific community” comprise … in controversial “Voiceprint” affairs? [Supported by International Association of Voice Identification.]

Voiceprirt Identification in the Courtroom

Abstract In a recent article appearing in the Journal of Forensic Sciences, J. J. Hennessey and C. H. Romig presented a review of the experiments involving identification by voice-print spectrograms, concluding that the various experiments in the field were conflicting, neither denying nor confirming the validity of voiceprint identification. This disagreement among scientists has had a tremendous affect on the admissibility of voiceprint identification as evidence in judicial proceedings. With one exception, until the end of 1970 evidence of voiceprint identification has been held inadmissible. Since that time, the results of an extensive experiment on voice identification conducted by Dr. Oscar Tosi, Professor of Audiology and Speech Sciences, Michigan State University, have been released and greatly strengthened the argument for the admissibility of voiceprint identification as competent evidence.

Effects of Context on Voiceprint Identification

This study investigated how well the file card system of Voiceprint Identification reported by Kersta (1962) fulfills its purpose of minimizing the effects of contextually caused spectral variation, and how well it serves as either a population reduction or identification tool. Subjects received training equivalent to that received by the experimenter in the Voiceprint Identification Training Course, and achieved error rates approximating those achieved by Kersta's subjects for similar training tasks. Experimental tasks required subjects to identify unknown speakers from a population of 50 known speakers by first eliminating all known speakers they were certain were not the unknown speaker, and then attempting identification. Ten attempts were made where the unknown and known speech samples of the same speaker were excerpted from the same phonetic contexts, and 10 attempts where they were excerpted from different contexts. Statistically significant differences in subject performance for same and different context tasks indicated that the file card system does not substantially minimize the effects of contextually caused spectral variation. The magnitude of error rates suggests that it cannot be used for either population reduction or identification.

Instruction and Application of Voiceprint Identification to Law Enforcement

The experience and success of Voiceprint Identification as applied to law-enforcement applications and the training of law-enforcement officers in the applied art are the dual purpose of this paper. The general method of application of Voiceprint Identification to law enforcement and the preparation of the exhibits for criminal trial presentation will be described. Examples drawn from the over 170 cases processed for over 35 law-enforcement agencies will be presented including conclusions for both probative and trial applications. A description of the Voiceprint Identification training course for qualified law-enforcement personnel and their success scores will be given. The two-week in-residence training has so far resulted in identification success scores exceeding 97% for this relatively short training period. Officers who were qualified fingerprint experts achieved success scores notably better than those without fingerprint-identification background.

Heartprints and Other Body Sounds

Analytical techniques used in voiceprint identification have been shown to give promise in the diagnosis of cardiovascular and other body sounds. This paper reports initial experimental observations in the diagnosis of aortic and pulmonic stenosis. Other potential uses are in the (1) determination of type and results of cleft-palate repair, (2) classification of voice changes occurring during asthmatic attacks, (3) estimation of the type and extent of eustachian-tube disease, (4) early detection of exacerbation of certain psychiatric (schizophrinic) states. The quantized spectral-pattern matching techniques used lend themselves to automated diagnosis as well as to teaching.

Voiceprint Identification

Initial successes indicated that a voiceprint identification system analogous to existing fingerprint identification is a possibility. Subjective matching of voiceprints has resulted in success scores of better than 97%. For the purposes of voiceprint identification, a library tape was recorded containing the voices of 15 male and 10 female, essentially normal American voices. The recording consisted of 10 cue words chosen for the high frequency with which they appear in telephone conversation. The cue words were uttered four times by each speaker, first singly, then in two five-word sentences, and finally in narrative during which each cue word appeared at least two times. Spectrograms were made from the above recordings and prepared for subjective testing. The voiceprints were randomly arranged for presentation in groups of 20, each group comprising four different utterances of a cue word by five different speakers. Upon initial presentation of five groups of 20 voiceprints to seven experienced subjects, four recorded perfect scores, two recorded one error, and one subject recorded two errors. A panel of naive female students under 18 years of age was chosen next and given five days of voiceprint reading experience. Upon presentation of the first voiceprint identification problem, the panel recorded only two errors in seven groups of 20 voiceprints In successive experiments, the matrices were expanded to 9 and then 12 talkers. Identification success scores exceeded 98% for over 5000 decisions. In a later series of experiments, simulated identification problems were investigated resulting in success scores exceeding 97%.