Line Spectral Frequencies Research Articles

Deep CNN for Parkinson’s Disease Classification Using Line Spectral Frequency Images of Sustained Speech Phonation

It is proposed to use Deep Convolution Neural Network (DCNN) which is a good classifier of natural images to learn speech spectrum images of sustained phonation to detect Parkinson’s Disease (PD) as an alternative to the existing feature-based machine learning method. It is shown that the proposed method yields very high accuracy without the need for separate feature computation stage. The speech spectrum representations proposed are Short Time Fourier Transform (STFT) spectrum of size Nx256 and Line Spectral Frequency (LSF) spectrum of size Nx16. LSF reflects the speech production mechanism and it is a novel idea to use LSF spectrum in DCNN to detect PD speech. The spectrum images look like random patterns and the performance is improved when using an additional deeper hidden layer of tampering pattern in the last stage of a fully connected layer. Using a standard PD-sustained phonation dataset the training accuracies achieved are 98.50% and 92.50% for STFT and LSF method, respectively. The validation accuracies achieved are 84.38% for STFT and 100% for LSF. The STFT method results in a sensitivity of 97.05%, a specificity of 88.63%, a precision of 86.84%, an F1-score of 91.66, a false positive rate (FPR) of 11.36%, and a false alarm rate of 12.82%. The LSF method results in a sensitivity 97.05%, a specificity of 95.45%, a precision of 94.28%, an F1-score of 95.65, an FPR of 4.50%, and a false alarm rate of 5.71%. The LSF based method performs better and the performance comparison with the state-of-the-art methods brings out the merits of the LSF spectrum image-based DCNN learning in PD detection using sustained phonation.

Frequency estimation of underwater acoustic targets based on improved FRI method

The line spectrum information of a ship’s radiated noise signal is often regarded as an important feature used to distinguish between targets. In this paper, an improved FRI (finite rate of innovation) method is proposed to estimate the low-frequency line spectra of the radiated noise from underwater acoustic passive targets under the condition of finite sampling points. First, a preliminary rough estimation of the frequency is performed using the Fourier transform. Then, the multichannel signal data are iteratively fitted based on the polynomial-ratio form model. The optimal estimation results are finally determined based on the residuals between the data and the model in the fitting process. Simulation experiments prove that compared with the original FRI algorithm, the method proposed in this paper can estimate line spectral frequency more efficiently under the condition of low signal-to-noise ratio (SNR), and can achieve fast convergence with shorter operation time. The experimental data processing results prove that the method can realize high-precision extraction of line spectrum information of ship noise.

A Machine Learning Integrated 5.8-GHz Continuous-Wave Radar for Honeybee Monitoring and Behavior Classification

A 5.8-GHz continuous-wave (CW) radar was developed and integrated on a compact printed circuit board (PCB) for near-hive monitoring of honeybees. It supported noninvasive detection of free-flying honeybees at 2 m and micro-Doppler extraction of bee wingbeat signatures at a closer range using both a double-balanced mixer and an in-phase/quadrature (IQ) mixer. An original technique combining full-wave simulations and Doppler-radar monitoring of pendulum motion was used for: radar calibration through spherical targets of different material and size; precise extraction of bee radar cross section (RCS) at 5.8 GHz; and estimation of detection range enhancement through partial silver coating of targets. Finally, the CW radar was integrated with machine learning (ML) to allow automated classification of incoming, outgoing, and hovering honeybees. Different ML approaches were tested, where the highest accuracy of 93.37% was found in ternary classification via support vector machine acting on line spectral frequencies.

Advancements in Speaker Recognition: Exploring Mel Frequency Cepstral Coefficients (MFCC) for Enhanced Performance in Speaker Recognition

Abstract: Speaker recognition, a fundamental capability of software or hardware systems, involves receiving speech signals, identifying the speaker present in the speech signal, and subsequently recognizing the speaker for future interactions. This process emulates the cognitive task performed by the human brain. At its core, speaker recognition begins with speech as the input to the system. Various techniques have been developed for speech recognition, including Mel frequency cepstral coefficients (MFCC), Linear Prediction Coefficients (LPC), Linear Prediction Cepstral coefficients (LPCC), Line Spectral Frequencies (LSF), Discrete Wavelet Transform (DWT), and Perceptual Linear Prediction (PLP). Although LPC and several other techniques have been explored, they are often deemed impractical for real-time applications. In contrast, MFCC stands out as one of the most prominent and widely used techniques for speaker recognition. The utilization of cepstrum allows for the computation of resemblance between two cepstral feature vectors, making it an effective tool in this domain. In comparison to LPC-derived cepstrum features, the use of MFCC features has demonstrated superior performance in metrics such as False Acceptance Rate (FAR) and False Rejection Rate (FRR) for speaker recognition systems. MFCCs leverage the human ear's critical bandwidth fluctuations with respect to frequency. To capture phonetically important characteristics of speech signals, filters are linearly separated at low frequencies and logarithmically separated at high frequencies. This design choice is central to the effectiveness of the MFCC technique. The primary objective of the proposed work is to devise efficient techniques that extract pertinent information related to the speaker, thereby enhancing the overall performance of the speaker recognition system. By optimizing feature extraction methods, this research aims to contribute to the advancement of speaker recognition technology.

Deep convolution neural network based Parkinson’s disease detection using line spectral frequency spectrum of running speech

The deformation of speech caused by glottic vocal tract is an early bio marker for Parkinson’s disease. A novel idea of Line Spectral Frequency trajectory spectrum image representation of the speech signals of the subjects in Deep Convolution Neural Network is proposed for Parkinson’s disease classification in which the convolution layer automatically learn the features from the input images and no separate feature calculation stage in required. The human vocal tract that produces a short phonetics is assumed as an all-pole Infinite impulse response system and the Line spectral frequency trajectory spectrum images represents the poles of the system and reflects the voice defects due to Parkinson’s disease. It is shown that the proposed method outperforms the existing state of the art work for two different utterance tasks one for sustained phonation and another for natural running speech dataset. It is demonstrated that the Deep Convolution Neural Network results in a training accuracy of 92.5% for sustained phonation dataset and training accuracy of 99.18% for King’s college running speech dataset. The validation accuracies for both the datasets are 100%. The proposed work is much better than another recent benchmark work in which Mel Frequency Cepstral Coefficient parameters are used in machine learning for Parkinson’s disease detection in running speech. The high performance of the proposed method for King’s college running speech dataset which is collected through mobile device voice recordings, gains attention. Rigorous performance analysis is performed for running speech dataset by using separate isolated test set for repeated 50 trials and the performance metrics are F1 score of 99.37%, sensitivity of 100%, precision of 98.75% and specificity of 99.27%.

Multiple time-instances features based approach for reference-free speech quality measurement

This paper investigates the problem of measuring speech quality of received speech signal without employing the original speech signal. The problem of deterioration of the speech quality arises due to noise present in the surroundings. To this line, we propose a multiple time-instances (MTI) features-based approach for reference-free speech quality measurement model. A voice activity detector (VAD) is exploited primarily for calculating the number of active speech chunks of a speech signal. For these chunks and their successive combinations called here batches, multi-resolution auditory model (MRAM), mel-frequency cepstral coefficients (MFCC) and line spectral frequencies (LSF) features are extracted and called as MTI features. It is hypothesized that the MTI features are capable in capturing the distortions caused by time-localized effects of short-time transients, impulsive noise, and its differences from the plosive sounds. The MTI metric estimates (MTI-ME) are calculated corresponding to these MTI features employing the Gaussian mixture model (GMM) probabilistic technique. The overall objective speech quality of a speech signal is then determined as a linear combination of optimally weighted MTI-ME corresponding to distinct active speech chunks and their successive combinations, that is, batches of that speech signal. Minimum mean square error criterion or Pearson’s correlation maximization criterion is employed for computing optimal weights. In addition, a deep neural network (DNN)-based speech quality model is also developed for calculating a single objective speech quality while considering all active speech chunks together. Pearson’s correlation coefficient and weighted average correlation are exploited for evaluating the performance. Results demonstrate that the proposed model achieves promising improvement over the standard speech quality model (P.563) and improves correlation values by around 37%.

Partial Discharge Detection of Multiple‐Narrow Band Antenna under the Radio Communication Noise

The ultra‐high frequency (UHF) method is attempted to diagnose the insulation condition of high voltage equipment as detecting partial discharge (PD) – emitted electromagnetic (EM) wave. Although the EM wave caused by PD has broadband, the UHF band includes many communication and broadcasting line spectral frequencies, which can affect PD detection as a noise. The authors have developed the multiple‐narrow band (MNB) antenna to detect discharge signals with high resistance against noisy line spectral frequencies. In this paper, the MNB antenna was applied to measure PD in SF6 gas and compared with a standard wide‐band Horn antenna to identify the PD detection ability. Furthermore, the MNB antenna was applied to measure PD signal in SF6 gas under the artificial and the real communication noise to estimate the noise elimination ability of MNB antenna. © 2021 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC.

Creating Song From Lip and Tongue Videos With a Convolutional Vocoder

A convolutional neural network and deep autoencoder are used to predict Line Spectral Frequencies, F0, and a voiced/unvoiced flag in singing data, using as input only ultrasound images of the tongue and visual images of the lips. A novel convolutional vocoder to transform the learned parameters into an audio signal is also presented. Spectral Distortion of predicted Line Spectral Frequencies is reduced compared to that in an earlier study using handcrafted features and multilayer perceptrons on the same data set; while predicted F0 and voiced/unvoiced flag predictions are found to be highly correlated with their ground truth values. Comparison of the convolutional vocoder to standard vocoders is made. Results can be of interest in the study of singing articulation as well as for silent speech interface research. Sample predicted audio files are available online. Source code: <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/TjuJianyu/SSI_DL</uri> .

A Multiple Narrow Band Antenna for Detection of Discharge‐Emitted Electromagnetic Waves for Insulation Diagnostic

Partial Discharge (PD) detection is an effective tool to diagnose the insulation condition of high voltage (HV) equipment. Among detection methods of PD, the ultra‐high frequency (UHF) method is distinguished by high sensitivity and safety operation. However, the UHF band includes many communication and broadcasting line spectral frequencies, which can be interference against the UHF method. From the point of view of the reduction of interferences, the authors proposed the multiple narrow band (MNB) antenna, which can detect PD signals with high sensitivity and eliminate the line spectral noise frequencies. In this paper, a new MNB antenna was evaluated based on fundamental properties and fabricated, then applied to detect discharge signals emitted from breakdown phenomena. The results demonstrated that the fabricated MNB antenna had a good omnidirectional radiation pattern and similar sensitivity with a standard wide‐band Horn antenna. © 2020 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC.

Evaluation of Potential Efficiency of Speech Coding Using Different Parameters of Linear Prediction

The paper presents the results for estimating the potential coding efficiency of spectrum envelope waveform (SEW) of speech signals (SS) by using the linear prediction (LP) method and using different sets of alternative equivalent parameters (AEP) that include line spectral pairs/projections (LSP), line spectral frequencies (LSF), and alternative line spectral parameters of highest splitting (LSP-HS and LSF-HS). These results are derived by the proposed approach based on using the LP method for SS SEW coding with maximum frame overlap during its analysis. It includes the examination of such coding scheme as approaching an appropriate analog vector source in each of AEP spaces, the stepwise designing of an appropriate vector codebook in each of spaces with gradual increase of its size and employing the ideal scheme of vector quantization with exhaustive search at each stage. The distortions versus rate relationships have been calculated based on the results of analysis in each of AEP spaces. In addition, a generalized function is proposed for approximation of the specified relationships. The technique is presented that makes it possible in each AEP space to estimate Shannon’s lower bound, the dispersion of the Gaussian equivalent source, differential entropy, redundancy, values of the weighting constant in the generalized formula of entropy, and other entropy characteristics of coding the equivalent sources (parameters of SS SEW) in these spaces. The efficiency indicators of real and potential coding of corresponding AEP are proposed and calculated. It has been shown that in conjunction with the combination of proposed efficiency indicators, the best results demonstrate spaces of line spectral parameters of highest splitting (LSP-HS and LSF-HS).

Single and multiple frame coding of LSF parameters using deep neural network and pyramid vector quantizer

In linear predictive speech coders, the linear predictive coding (LPC) parameters are usually transformed into the line spectral frequency (LSF) representation for quantization. In this paper, the single and multiple frame coding of LSF parameters using deep neural network (DNN) and pyramid vector quanrizer (PVQ) are proposed. In the single-frame scheme, a non-linear DNN predictor which demonstrates much better prediction performance than the autoregressive (AR) model is applied to exploit the inter-frame dependency of LSF parameters. The prediction residual signal has Laplacian distribution and can be efficiently quantized by the PVQ. The performance evaluation using spectral distortion shows that the proposed DNN predictive PVQ outperforms the AR predictive split vector quantization (SVQ). In the multiple-frame scheme, a deep autoencoder possessing linear coder-layer units with Gaussian noise is used to compress and de-correlate multiple LSF frames. The deep autoencoder shows a high degree of modelling flexibility for multiple LSF frames. To quantize the coder-layer vector effectively, a PVQ is considered. The experimental results show that the proposed multi-frame scheme with determined optimal coder-layer dimension outperforms the discrete cosine model (DCM)-based approach in terms of spectral distortion performance and robustness across different speech segments.

Speaker-independent expressive voice synthesis using learning-based hybrid network model

Emotional voice conversion systems are used to formulate mapping functions to transform the neutral speech from output of text-to-speech systems to that of target emotion appropriate to the context. In this work, a learning-based hybrid model is proposed for speaker-independent emotional voice conversion using a combination of deep belief nets (DBN-DNN) and general regression neural net (GRNN). The main acoustic features considered for mapping are shape of the vocal tract given by line spectral frequencies (LSF), glottal excitation given by LP residual and long term prosodic features viz. pitch contour and energy. GRNN is used to attain the transformation function between source and target LSFs. Source and target LP residuals are subjected to wavelet transform before DBN-DNN training. This is helpful to remove phase-change induced distortions which may affect the performance of neural networks when transforming time-domain residual. Low-dimensional pitch (intonation) contour is subjected to feed-forward neural network mapping (ANN). Energy modification is achieved by taking average transformation scales across entire utterance. The system is tested on three different datasets viz. EmoDB (German), IITKGP (Telugu) and SAVEE (English). Relative performances of proposed model are compared with constrained variance GMM (CV-GMM) using objective and subjective metrics. The results obtained show a significant performance improvement of 41% in RMSE (Hz) and 9.72% in Pearson’s correlation coefficient for fundamental frequency (F0) (Fear) compared to CV-GMM across all 3 datasets. Subjective results indicate a maximum MOS score of 3.85 (Fear) and CMOS score of 3.9 (Happiness) across the three datasets considered.

Speech Watermarking for Tampering Detection Based on Modifications to LSFs

There have been serious issues concerning the protection of speech signals from malicious tampering. Digital watermarking has been paid much attention in solving this problem. This paper proposes a tampering detection approach based on speech watermarking by modifying the line spectral frequencies (LSFs). Watermarks are embedded into LSFs that derived from linear prediction (LP) analysis with dither modulation-quantization index modulation (DM-QIM). Minor modifications to LSFs introduced by quantization not only enable the watermarks to be inaudible to human auditory system but also provide the possibility of robustness against meaningful processing and fragility against tampering. We evaluated the proposed approach with objective evaluations with respect to inaudibility, robustness, and fragility. The results indicated that the proposed approach for tampering detection not only satisfied inaudibility but also provided good robustness against meaningful processing and fragility against malicious tampering.

Single-ended Speech Quality Evaluation Using Linear Combination of the Quality Score Estimates of Multi-instances Features

Background: In a single-ended speech quality evaluation, the measurement of Mean Opinion Score (MOS) is done objectively without the use of clean speech as a reference. In this work, multiple time-instances or multi-instances features using Multi-Resolution Auditory Model (MRAM) along with other relevant features such as Mel-Frequency Cepstral Coefficients (MFCC), and Line Spectral Frequencies (LSF) features are used for single-ended speech quality measurement. The Voice Activity Detection (VAD) algorithm separates the presence of speech-regions from silence in a speech signal. Methods: The multi-instances features are computed using MRAM, MFCC and LSF for different combinations of speech-regions to capture degradations due to multiple time-localized effects or the attacks of short-time transient distortions such as impulsive noise and their distinctions from plosive sounds in speech. These multi-instances features are used for Gaussian Mixture Model (GMM) MAPPING to compute the objective MOS values, corresponding to all multi-instances features. Results: The overall objective MOS estimation of the speech signal is calculated by averaging all the values of objective MOS corresponding to the different multi-instances features of a speech signal. The results in terms of Pearson’s Correlation Coefficients (PCC) and root mean square error (RMSE) between the subjective MOS and the estimated overall objective MOS of speech signals are computed and compared with International Telecommunication Union-telephony (ITU-T) Recommendation P.563 and recently published works on similar types of databases. Conclusion: The improved values of PCC and RMSE between the subjective and the estimated overall objective MOS show the efficacy of the approach.

Design and Comparison of Vector Quantization Codebooks for Narrowband Speech Coding

Vector quantization codebook algorithms are used for coding of narrow band speech signals. Multi-stage vector quantization and split vector quantization methods are two important techniques used for coding of narrowband speech signals and these methods are very popular due to the high bit rate minimization during coding of the signals. This paper presents performance measurements of multistage vector quantization and split vector quantization methods. We used line spectral frequencies for coding of the speech signals in codebook tables so as to ensure filter stability after quantization. The codebooks were generated by using the Linde-Buzo-Gray (LBG) algorithm. The tests were performed by selecting large amount of input data in training and test stages and to evaluate noise robustness of the methods, both noisy and clean speech signals were used. As a result, different codebooks were designed and tested in many stages and different bit rates to measure quantization performance. It is measured in terms of spectral distortion evaluation. We obtained the best result in 24bit multistage vector quantization codebook with a spectral distortion less than 1 dB for clean speech training data input. When we compared multistage and split vector quantization codebook spectral distortion results, multistage codebooks gave better performance in each option.

A steganographic method based on gain quantization for iLBC speech streams

Under the premise of ensuring good speech quality and resistance to steganalysis, how to embed as much information as possible into low-bit-rate speech is a challenge. The existing research mainly focuses on least significant bits in the compressed speech streams or line spectral frequencies in the encoding processes of the G.723.1 and G.729a codecs. In this paper, we are concentrating on steganography in Internet Low Bitrate Codec speech streams. Specifically, we present a novel steganographic method on the basis of the gain quantization (GQS), which is included in the vector quantization of dynamic codebooks. The secret information is embedded by modifying the search range of the gain codebook to quantize the gain value. The experimental results show that the proposed GQS method achieves a higher hiding capacity, less signal distortion, and better resistance to steganalysis.

Detection of speech tampering using sparse representations and spectral manipulations based information hiding

Speech tampering has brought serious problems to speech security. Information hiding method can be used for tampering detection if it can satisfy several competitive requirements, e.g., inaudibility, robustness, blindness, and fragility. According to preliminary analysis, spectral envelope and formants are important indicators of tampering, since tampering the speech will unavoidably modify the shape of the spectral envelope and the locations/magnitudes of the formants. By taking advantage of this, this paper proposes a spectral manipulations based information hiding method for tampering detection in sparse domain. To robustly extract the embedded information, the Robust Principal Component Analysis (RPCA) is employed to decompose the original speech into sparse component and low-rank component. The sparse component which contains the main spectral envelope and formant structure is selected for information hiding/embedding via spectral manipulations, by controlling the shape and power of formants with line spectral frequencies (LSFs). Evaluation results suggest that the proposed method can satisfy inaudibility, robustness, and fragility. Furthermore, it is able to detect both the temporal tampering and acoustic feature based tampering with reasonable accuracy.

Towards Security of GSM Voice Communication

Global system for mobile communication (GSM) is widely used digital mobile service around the world. Although GSM was designed as a secure wireless system, it is now vulnerable to different targeted attacks. There is a need to address security domains especially the confidentiality of communication. This paper presents a novel framework for end to end secure voice communication over the GSM networks using encryption algorithm AES-256. A special Modem and speech coding technique are designed to enable the transmission of encrypted speech using GSM voice channel. To the best of our knowledge, this is first solution that uses single codebook for transmission of secure voice. An efficient low bit-rate (1.9 kbps) speech coder is also designed for use with the proposed modulation scheme for optimal results. Different speech characteristics such as pitch, energy and line spectral frequencies are extracted and preserved before compression and encryption of speech. Previously, the best achieved data rate was 1.6 kbps with three codebooks, whilst the proposed approach achieves 2 kbps with 0% bit error rate. The empirical results show that the methodology can be used for real time applications to transmit encrypted voice using GSM network.

Methods for Calculating and Coding the Parameters of Autoregressive Speech Model when Developing the Vocoder Based on Fixed Point Signal Process

Consideration is given to calculating and coding the parameters of vocoders operating at speed of 1200 and 2400 bps. The efficient methods for calculating and coding the parameters of the autoregressive speech signal model such as line spectral frequencies and the pitch period are proposed. The works on computation optimization allowed a vocoder to be implemented in real time on the signal processor ADSP-2191.

Improving the Estimation of Sinusoidal Frequencies and Direction-of-Arrival Using Line Spectral Frequencies

In this letter we address the classic problem of estimating the frequencies of multiple sinusoids in the presence of noise. We use line spectral frequencies (LSFs) and introduce the idea of “closeness of LSFs” and propose two methods based on it. The minimum phase annihilating filter $\boldsymbol {A(z)}$ is estimated using MUSIC. The corresponding LSFs are then processed to estimate the frequencies, as opposed to estimating them from the roots that are closest to the unit circle. The closeness of LSFs criterion leads to a lower threshold when compared to MUSIC, with improvements ranging from $\text{0}$ to $\text{2}$ dB for the well-known two-sinusoids example; with pseudo-noise resampling the improvement is up to $\text{4}$ dB. When applied to direction of arrival estimation, for the example considered by Shaghaghi and Vorobyov, the threshold improves by $\text{6}$ dB and matches that given by the stochastic ML method.