Speech Compression Research Articles

An Efficient Codebook Search Algorithm for Line Spectrum Frequency (LSF) Vector Quantization in Speech Codec

A high-performance vector quantization (VQ) codebook search algorithm is proposed in this paper. VQ is an important data compression technique that has been widely applied to speech, image, and video compression. However, the process of the codebook search demands a high computational load. To solve this issue, a novel algorithm that consists of training and encoding procedures is proposed. In the training procedure, a training speech dataset was used to build the squared-error distortion look-up table for each subspace. In the encoding procedure, firstly, an input vector was quickly assigned to a search subspace. Secondly, the candidate code word group was obtained by employing the triangular inequality elimination (TIE) equation. Finally, a partial distortion elimination technique was employed to reduce the number of multiplications. The proposed method reduced the number of searches and computation load significantly, especially when the input vectors were uncorrelated. The experimental results show that the proposed algorithm provides a computational saving (CS) of up to 85% in the full search algorithm, up to 76% in the TIE algorithm, and up to 63% in the iterative TIE algorithm. Further, the proposed method provides CS and load reduction of up to 29–33% and 67–69%, respectively, over the BSS-ITIE algorithm.

Speech compression and encryption based on discrete wavelet transform and chaotic signals

To increase transfer and storage efficiencies of the information, data compression has emerged as a significant issue in the communication environments. This paper introduces compression and encryption of speech signals based on Discrete Wavelet Transform (DWT) and Chaotic signals. DWT sparsens and codes the speech signal to the wavelet coefficients. The less impactful coefficients are eliminated to reduce the amount of data. After that, a new coding process which utilizes the chaotic signals is proposed to encode, in encrypted form, the residual coefficients. A High strength to the encryption process is realized by using four linked Henon Chaotic Maps (HCM) in the proposed scheme. Multi HCM guarantees larger than 10240 of key space to the encryption process. The proposed system obtains up to −41.449 dB of spectral segmental signal-to-noise ratio, which measures and proves the strength of encryption. Also, at 10% compression ratio, signal-to-noise ratio of 11.549 dB and perceptual evaluation speech quality of 3.02945 demonstrate that the proposed system has high quality and intelligibility of the reconstructed speech.

General Steganalysis Method of Compressed Speech Under Different Standards

Analysis-by-synthesis linear predictive coding (AbS-LPC) is widely used in a variety of low-bit-rate speech codecs. Most of the current steganalysis methods for AbS-LPC low-bit-rate compressed speech steganography are specifically designed for a specific coding standard or category of steganography methods, and thus lack generalization capability. In this paper, a general steganalysis method for detecting steganographies in low-bit-rate compressed speech under different standards is proposed. First, the code-element matrices corresponding to different coding standards are concatenated to obtain a synthetic code-element matrix, which will be mapped into an intermediate feature representation by utilizing the pre-trained dictionaries. Then, bidirectional long short-term memory is employed to capture long-term contextual correlations. Finally, a code-element affinity attention mechanism is used to capture the global inter-frame context, and a full connection structure is used to generate the prediction result. Experimental results show that the proposed method is effective and better than the comparison methods for detecting steganographies in cross-standard low-bit-rate compressed speech.

Detection of Multiple Steganography Methods in Compressed Speech Based on Code Element Embedding, Bi-LSTM and CNN With Attention Mechanisms

Steganographic algorithms in low-bit-rate compressed speech bring convenience to realize covert communication, meanwhile result in safety issues. The existing steganalysis methods are normally designed for one specific category of steganographic methods, thus lacking generalization capability. In this paper, we propose a general steganalysis method based on code element (CE) embedding, Bi-LSTM and CNN with attention mechanisms. Firstly, CEs in each frame are converted to a multi-hot vector. And each multi-hot vector will be mapped into a fixed-length embedding vector to get a more compact representation by utilizing dictionaries. Then, Bi-LSTM and CNN are applied to extract the contextual information and the local characteristics respectively of these embedding vectors. In addition, the attention mechanisms are introduced in different layers of the network to assign different weights to the output feature within each layer. Finally, the prediction results can be generated by the fully connected layer. Experimental results show that our method performs better than the existing steganalysis methods for detecting multiple steganography methods in the low-bit-rate compressed speech streams.

Real Time Implementation of Voice Activity Detection based on False Acceptance Regulation

In speech compression systems, Voice Activity Detection (VAD) is frequently used to distinguish active voice from other noisy sounds. In this paper, a robust approach of VAD is presented to deal with non-stationary noisy environments. The proposed algorithm exploits adaptive thresholding technique to keep a desired False Acceptance (FA) rate. Iterative hypothesis tests, using signal energy, are implemented to discard or to accept the successive audio frames as active voice. According to the stationary property of the speech, we provide a smoothing method to obtain final VAD decisions. The main contribution of the proposed algorithm concerns its ability to automatically adjust the energy threshold according to the local noise estimator. We analyzed the proposed approach by presenting a comparison with the G.729-B via the NOIZEUS database. The VAD architecture is implemented on a Microcontroller-based system (MCU). Several tests have been conducted by performing real time acquisition via the Input/Output ports of the MCU-system.

Study on Speech Compression and Encryption for Signal Processing

Study on Speech Compression and Encryption for Signal Processing

Parkinson’s Diagnosis through Compressed Speech Signals at remote Locations


 
 
 Rapid increase in Parkinson’s disease all over the world before the age of 60’s is alarming. Parkinson’s affects the neurological controls of human body. In this paper we intend to adopt machine based learning to automate the diagnosis of Parkinson’s disease with help of compressed speech signals sent via network. An advanced Solution is proposed to automate the diagnosis process more rapid than that of ordinary process. In this study Speech samples from 55 subjects have been collected, 43 with Parkinson Disease and 12 Healthy subjects. Speech signals from clients captured from various sensors and devices transmitted to the cloud for processing. In the cloud Speech samples are compressed using compression codecs MP3, MP4, G.722, G.226, GSM-EFR, AMR-WB, SVOPC/SILK, and OPUS, and then same are diagnosed without compression impact and deliver accurate results. Three classifiers SVM (Support Vector Machine), NN (Neural Network), and GA (Genetic Algorithm) are applied and calculated accurate results as compared with compressed and un-compressed voice samples. Successful experiments enabled us to achieve 90.7% accuracy. Same process can be used for online and social media applications which are available in portable devices as well.
 
 

Does Time Compression Decrease Intelligibility for Female Talkers More Than for Male Talkers?

Purpose This preliminary investigation compared effects of time compression on intelligibility for male versus female talkers. We hypothesized that time compression would have a greater effect for female talkers. Method Sentence materials from four talkers (two males) were time compressed, and original-speed and time-compressed speech materials were presented in a background of 12-talker babble to young adult listeners with normal hearing. Each talker/processing condition was heard by eight listeners (total N = 64). Generalized linear mixed-effects models were used to determine the effects of and interaction between processing condition and talker sex on keyword intelligibility. Additional post hoc analyses examined whether processing condition effects were related to talker vowel space and word frequency. Results For original-speed sentences, female and male talkers were essentially equally intelligible. Time compression reduced intelligibility for all talkers, but the effect was significantly greater for the female talkers. Supplementary analyses revealed that the effect of time compression depended on both talker vowel space and word frequency: The detrimental effect decreased significantly as word frequency and vowel space increased. Word frequency effects were also greater overall for talkers with larger vowel spaces. Conclusions While the small talker sample limits conclusions about the effects of talker sex, the secondary analyses suggest that intelligibility of talkers with larger vowel spaces is less susceptible to the negative effect of time compression, especially for high-frequency words.

Tasks, Talkers, and the Perceptual Learning of Time-Compressed Speech

ABSTRACT Many auditory skills (e.g., the recognition of perceptually difficult speech) improve with practice, but only limited transfer of this learning to untrained materials was observed in previous studies. Here, we asked whether the type of practice (semantic or accent judgment) and talker variability (defined as the number of different talkers encountered during practice, two or six), influenced the perceptual learning of time-compressed speech and its transfer to unpracticed materials. Four groups of participants trained on the four task/talker variability combinations, and their pre- and post-training recognition of time-compressed speech was compared to that of a group of untrained participants (n = 14–16 participants/group). Across groups, training led to substantial learning of the trained tokens and to transfer to new talkers producing previously encountered sentences (compared to the untrained control group). However, neither type of training had a significant effect on the recognition of new sentences, even for familiar talkers. Semantic training yielded more learning and better retention of learning over a 2-week interval than accent training. The number of talkers had only marginal effects. These results suggest that learning of time-compressed speech is robust and only partially task-specific, but its transfer to untrained tokens following brief practice is limited. In contrast to other types of speech training, here exposure to a variety of talkers during training did not contribute to the transfer of learning to new materials.

AN EFFICIENT SPEECH GENERATIVE MODEL BASED ON DETERMINISTIC/STOCHASTIC SEPARATION OF SPECTRAL ENVELOPES

The paper presents a speech generative model that provides an efficient way of generating speech waveform from its amplitude spectral envelopes. The model is based on hybrid speech representation that includes deterministic (harmonic) and stochastic (noise) components. The main idea behind the approach originates from the fact that speech signal has a determined spectral structure that is statistically bound with deterministic/stochastic energy distribution in the spectrum. The performance of the model is evaluated using an experimental low-bitrate wide-band speech coder. The quality of reconstructed speech is evaluated using objective and subjective methods. Two objective quality characteristics were calculated: Modified Bark Spectral Distortion (MBSD) and Perceptual Evaluation of Speech Quality (PESQ). Narrow-band and wide-band versions of the proposed solution were compared with MELP (Mixed Excitation Linear Prediction) speech coder and AMR (Adaptive Multi-Rate) speech coder, respectively. The speech base of two female and two male speakers were used for testing. The performed tests show that overall performance of the proposed approach is speaker-dependent and it is better for male voices. Supposedly, this difference indicates the influence of pitch highness on separation accuracy. In that way, using the proposed approach in experimental speech compression system provides decent MBSD values and comparable PESQ values with AMR speech coder at 6,6 kbit/s. Additional subjective listening testsdemonstrate that the implemented coding system retains phonetic content and speaker’s identity. It proves consistency of the proposed approach.

Research on Application of Digital Signal Processing Technology in Communication

With the continuous development of modern information technology and the continuous improvement of people’s living standards, electronic information technology has been widely used in people’s lives. People’s lives are inseparable from it. With its continuous development, its application range is getting wider and wider. This article first gives a brief overview of digital signal processing technology, analyzes the advantages and disadvantages of digital signal processing technology in the communication field, and the specific applications of speech compression coding and software radio in communication are analyzed in order to better promote the application of digital signal processing technology in the communication field.

Time-compressed speech test in adults with and without central auditory processing disorders

ABSTRACT Purpose: to analyze and compare the performance in the time-compressed speech test and the auditory behavior of adults with and without central auditory processing disorders. Methods: an observational, analytical, cross-sectional study with a total of 40 people of both genders aged 18 to 35 years participating in the study. They were submitted to anamnesis, basic audiological assessment, and a core battery of tests for central auditory processing - including the dichotic digits test (binaural integration), frequency pattern test, and time-compressed speech test (TCST). Based on the results of the dichotic digits and frequency pattern tests, the subjects were divided into two groups, with and without central auditory processing disorders. The auditory behavior was assessed with the Scale of Auditory Behavior (SAB) questionnaire. The Mann-Whitney and Fisher’s exact tests were used for the statistical analysis, setting the significance level at p < 0.05. Results: no difference in performance was found between the groups regarding the ears. There was a difference between the groups only in the time-compressed speech test with monosyllable stimuli in the left ear (p = 0.026). Monosyllables were the words that resulted in most errors. Conclusion: it was verified that only the list of stimuli influenced the performance, differing the individuals with and without central auditory processing disorders. There was an association of auditory behavior, analyzed with the SAB questionnaire, with the performance in the TCST with the list of monosyllables. It is suggested that this list be used when assessing adults by the time-compressed speech test.

Behavioral tests in adults: reference values and comparison between groups presenting or not central auditory processing disorder

ABSTRACT Purpose: to generate reference values for different behavioral central auditory processing (CAP) tests in normally hearing adults, with age stratification, as well as to compare the results, according to the presence or absence of CAP disorder (CAPD). Methods: the study sample consisted of 94 adults with normal auditory thresholds, with at least 11 years of schooling, who were divided into two groups, according to the performance in the Dichotic Sentences Identification Test (DSI), Masking Level Difference (MLD), Random Gap Detection Test (RGDT), Adapted Time-Compressed Speech (ATCS) and Duration and Frequency Pattern Tests (DPT/FPT).The 64 subjects who presented normality in all the tests formed the G1 group and the 30 subjects who presented alteration in at least one of them, originated the G2 group. To accomplish the analysis it was used the Mann-Whitney U Test. In all analyzes, the significance level was 5% (p≤0.05). Results: the sum of the two standard deviations for the mean performance of G1 yielded reference values for the different tests addressed. When G1 and G2 groups were compared, G1 presented better results, this being significant in the DSI (left ear), MLD, RGDT, DPT and FPT tests, for the group aged between 18 and 29 years and, DSI (right ear), RGDT and ATCS (right ear), for the group from 30 to 58 years. Conclusion: it was possible to generate reference values for different behavioral tests of CAP in normally hearing adults, as well as to verify a better performance for the group with no CAPD.

Modulation masking release reduction as a function of time-compressed speech

ABSTRACT Purpose: to investigate the magnitude of the modulation masking release in sentence recognition as a function of compression level and modulation rate. Methods: sentences of the Brazilian Portuguese version of the Hearing in Noise Test sentences were used as stimulus. The sentence recognition thresholds were established as a function of speech compression level (0%, 33%, and 50%) in steady and modulated noise at different modulation rates (4, 10, 32 Hz). The analysis of variance was performed for repeated measures, using the 5% significance level. Results: sentence recognition thresholds were higher for higher compression levels in the different types of noise. However, thresholds were smaller for modulated noises. Also, the magnitude of modulation masking release decreased as speech compression level increased. Nevertheless, no difference was observed in compressed speech between different noise modulation rates, in relation to the speech compression level. Conclusion: the magnitude of the modulation masking release decreased as the speech time-compression increased. Also, the reductions in modulation masking release, in relation to the speech time-compression level, did not differ between the masking-noise modulation rates (4, 10, and 32 Hz).

Rapid Perceptual Learning: A Potential Source of Individual Differences in Speech Perception Under Adverse Conditions?

Challenging listening situations (e.g., when speech is rapid or noisy) result in substantial individual differences in speech perception. We propose that rapid auditory perceptual learning is one of the factors contributing to those individual differences. To explore this proposal, we assessed rapid perceptual learning of time-compressed speech in young adults with normal hearing and in older adults with age-related hearing loss. We also assessed the contribution of this learning as well as that of hearing and cognition (vocabulary, working memory, and selective attention) to the recognition of natural-fast speech (NFS; both groups) and speech in noise (younger adults). In young adults, rapid learning and vocabulary were significant predictors of NFS and speech in noise recognition. In older adults, hearing thresholds, vocabulary, and rapid learning were significant predictors of NFS recognition. In both groups, models that included learning fitted the speech data better than models that did not include learning. Therefore, under adverse conditions, rapid learning may be one of the skills listeners could employ to support speech recognition.

Research on Application of Digital Signal Processing Technology in Communication

With the continuous development of modern information technology and the continuous improvement of people's living standards, electronic information technology has been widely used in people's lives. People's lives are inseparable from it. With its continuous development, its application range is getting wider and wider. This article first gives a brief overview of digital signal processing technology, analyzes the advantages and disadvantages of digital signal processing technology in the communication field, and the specific applications of speech compression coding and software radio in communication are analyzed in order to better promote the application of digital signal processing technology in the communication field.

New Speech Compression Technique based on Filter Bank Design and Psychoacoustic Model

In this paper, a new speech compression technique is proposed. This technique applies a Psychoacoustic Model and a general approach for Filter Bank Design using optimization. It is evaluated and compared with a compression technique using a MDCT (Modified Discrete Cosine Transform) Filter Bank of 32 Filters and a Psychoacoustic Model. This evaluation and comparison is performed by calculating bits before and after compression, PSNR (Peak Signal to Noise Ratio), NRMSE (Normalized Root Mean Square Error), SNR (Signal to Noise Ratio) and PESQ (Perceptual evaluation of speech quality) computations. The two techniques are tested and applied to a number of speech signals that are sampled at 8 kHz. The results obtained from this evaluation show that the proposed technique outperforms the second compression technique (based on a Psychoacoustic Model and MDCT filter Bank) in terms of Bits after compression and compression ratio. In fact, the proposed technique yields higher values for the compression ratio than the second compression technique. Moreover, the proposed compression technique presents reconstructed speech signals with acceptable perceptual qualities. This is justified by the values of SNR, PSNR and NRMSE and PESQ.

Monitoring vital signs with time-compressed speech.

Spearcons-time-compressed speech phrases-may be an effective way of communicating vital signs to clinicians without disturbing patients and their families. Four experiments tested the effectiveness of spearcons for conveying oxygen saturation (SpO2) and heart rate (HR) of one or more patients. Experiment 1 demonstrated that spearcons were more effective than earcons (abstract auditory motifs) at conveying clinical ranges. Experiment 2 demonstrated that casual listeners could not learn to decipher the spearcons whereas listeners told the exact vocabulary could. Experiment 3 demonstrated that participants could interpret sequences of sounds representing multiple patients better with spearcons than with pitch-based earcons, especially when tones replaced the spearcons for normal patients. Experiment 4 compared multiple-patient monitoring of two vital signs with either spearcons, a visual display showing SpO2 and HR in the same temporal sequence as the spearcons, or a visual display showing multiple patient levels simultaneously. All displays conveyed which patients were abnormal with high accuracy. Visual displays better conveyed the vital sign levels for each patient, but cannot be used eyes-free. All displays showed accuracy decrements with working memory load. Spearcons may be viable for single and multiple patient monitoring. Further research should test spearcons with more vital signs, during multitasking, and longitudinally. (PsycINFO Database Record (c) 2019 APA, all rights reserved).

The Impact of Concurrent Linguistic Tasks on Participants’ Identification of Spearcons

Spearcons (time-compressed speech) may be a viable display for patient monitoring, but the impact of concurrent linguistic tasks on spearcons has not been examined. We tested whether different concurrent linguistic tasks worsen participants’ identification of spearcons. Experiment 1 tested participants’ identification of spearcons representing 2 vital signs of 5 patients while participants did either no concurrent task, reading, or saying linguistic tasks. Experiment 2 tested identification of 48 single patient-monitoring spearcons while participants did no concurrent task, reading, listening, and saying linguistic tasks. In Experiment 1 the saying task worsened participants’ identification of spearcons compared with the other tasks. In Experiment 2, the saying and listening tasks each reduced participants’ accuracy at identifying spearcons, but the reading task did not. Listening had no more effect than the saying task did. Concurrent listening and saying tasks worsen participants’ identification of spearcons, probably due to auditory modality interference in working memory.

Spearcons for Patient Monitoring: Program of Laboratory-Based Feasibility Studies

Clinicians are not always at their patients’ bedsides and may therefore need ways of remotely monitoring the well-being of multiple patients under their care. We outline the main findings of a research program investigating whether the intermittent presentation of short phrases of time-compressed speech (spearcons) is an effective way of giving mobile clinicians information about their patients without annoying either clinician or patient. We provide a high-level overview of several studies investigating participants’ ability to understand spearcons, both individually and in sequences representing multiple patients. We then report in more detail a recent small study testing whether participants’ ability to understand spearcons is compromised by different kinds of ongoing tasks. Finally, we outline further issues that should be addressed and further research studies performed before spearcons could be considered a viable tool for patient monitoring.