Speech Files Research Articles

Spectral Subtraction Based Weighted Function for Pitch Extraction in Noisy Speech

Many speech-related works employ the pitch period as a crucial component. Speech signals are typically collected in challenging noisy settings for real-world projects. Therefore, it is now more important than ever for the algorithm to be noise resistant in order to estimate pitch accurately. However, when dealing with noisy speech files at a low signal-to-noise ratio (SNR) value, many state-of-the-art algorithms are unable to produce satisfactory results. In this work, a new noise-resistant pitch estimation algorithm based on spectral subtraction is presented, which uses a weighted function to lessen the impact of the vocal tract effect. Furthermore, to enhance the correlation between the pitch estimates and smoothen the pitch contours, we employ a weighted function that combines the spectral subtraction-based technique as the numerator and the circular average magnitude difference function (CAMDF) as the denominator. We have utilized two noisy speech databases using seven different kinds of recorded ambient noise, and we evaluated our system against three cutting-edge methods. The suggested method lowers the Gross Pitch Error (GPE) rate at practically all SNRs in white noise and performs best on the NTT and KEELE databases.

Cross-corpus speech emotion recognition with transformers: Leveraging handcrafted features and data augmentation

Speech emotion recognition (SER) stands as a prominent and dynamic research field in data science due to its extensive application in various domains such as psychological assessment, mobile services, and computer games, mobile services. In previous research, numerous studies utilized manually engineered features for emotion classification, resulting in commendable accuracy. However, these features tend to underperform in complex scenarios, leading to reduced classification accuracy. These scenarios include: 1. Datasets that contain diverse speech patterns, dialects, accents, or variations in emotional expressions. 2. Data with background noise. 3. Scenarios where the distribution of emotions varies significantly across datasets can be challenging. 4. Combining datasets from different sources introduce complexities due to variations in recording conditions, data quality, and emotional expressions. Consequently, there is a need to improve the classification performance of SER techniques. To address this, a novel SER framework was introduced in this study. Prior to feature extraction, signal preprocessing and data augmentation methods were applied to augment the available data, resulting in the derivation of 18 informative features from each signal. The discriminative feature set was obtained using feature selection techniques which was then utilized as input for emotion recognition using the SAVEE, RAVDESS, and EMO-DB datasets. Furthermore, this research also implemented a cross-corpus model that incorporated all speech files related to common emotions from three datasets. The experimental outcomes demonstrated the superior performance of SER framework compared to existing frameworks in the field. Notably, the framework presented in this study achieved remarkable accuracy rates across various datasets. Specifically, the proposed model obtained an accuracy of 95%, 94%,97%, and 97% on SAVEE, RAVDESS, EMO-DB and cross-corpus datasets respectively. These results underscore the significant contribution of our proposed framework to the field of SER.

Pathological voice classification using MEEL features and SVM-TabNet model

In clinical settings, early diagnosis and objective assessment depend on the detection of voice pathology. To classify anomalous voices, this work uses an approach that combines the SVM-TabNet fusion model with MEEL (Mel-Frequency Energy Line) features. Further, the dataset consists of 1037 speech files, including recordings from people with laryngocele and Vox senilis as well as from healthy persons. Additionally, the main goal is to create an efficient classification model that can differentiate between normal and abnormal voice patterns. Modern techniques frequently lack the accuracy required for a precise diagnosis, which highlights the need for novel strategies. The suggested approach uses an SVM-TabNet fusion model for classification after feature extraction using MEEL characteristics. MEEL features provide extensive information for categorization by capturing complex patterns in audio transmissions. Moreover, by combining the advantages of SVM and TabNet models, classification performance is improved. Moreover, testing the model on test data yields remarkable results: 99.7 % accuracy, 0.992 F1 score, 0.996 precision, and 0.995 recall. Additional testing on additional datasets reliably validates outstanding performance, with 99.4 % accuracy, 0.99 F1 score, 0.998 precision, and 0.989 % recall. Furthermore, using the Saarbruecken Voice Database (SVD), the suggested methodology achieves an impressive accuracy of 99.97 %, demonstrating its durability and generalizability across many datasets. Overall, this work shows how the SVM-TabNet fusion model with MEEL characteristics may be used to accurately and consistently classify diseased voices, providing encouraging opportunities for clinical diagnosis and therapy tracking.

Secure Stego Method to Protect SMS Transmission

A novel method of message steganography will be proposed. The method will use a digital speech file as a covering media. The 32 binary bits of each speech sample will be used for data hiding, using these bits will increase the hiding capacity, and changing them will not much affect the sample value, thus the quality of the stego file will be always high. The proposed method will use a private key to protect the hidden message, the protection will be applied by shuffling the rows of the speech binary matrix using the contents of the generated indices key, The induces key will be generated by running a chaotic logistic map model using the values of the chaotic parameters included in the private key. The proposed method will be tested and implemented using various messages, the quality, sensitivity and speed of the proposed method will be analyzed to prove the enhancements provided by the proposed method.

Enhanced voice over 5G transmission with unequal error protection and generative adversarial networks

SummaryVoice services have progressed significantly since the launch of mobile phones and have evolved from analog technology to digital ones. Since the launch of 5G, voice over 5G (Vo5G) will slowly take over legacy mobile technologies to provide better voice services. Nevertheless, interruptions and noise will continue to affect speech communication over wireless networks, ensuing in a reduced Quality of Service (QoS). Three enhanced transmission schemes for Vo5G are proposed. The first proposal benefits from the difference in weight of the bits of internet Low Bitrate Code (iLBC) coded packets to offer them diverse degrees of protection i.e. Unequal Error Protection (UEP). UEP enhances the reception quality by allowing the more significant bits to be decoded with greater reliability during 5G transmission. The second and third schemes substitute the usual Quadrature Amplitude Modulation (QAM) constellation with a Statistical QAM (SQAM) one. The second scheme arranges the significant bits on the SQAM constellation and the remaining bits onto the conventional QAM arrangement. For the third scheme, all the bits are arranged on the SQAM constellation. Generative Adversarial Networks (GAN) were incorporated into the third scheme as the third scheme provided better gains over the first and second schemes. GAN was used to synthesize and enhance parts of the speech file that were lost during transmission over the 5G channel resulting in a gain in Segmented Signal to Noise Ratio (SSNR). The addition of GAN boosted the gain obtained by the third scheme by 1.45 dB with 16‐QAM at a code rate of 1/3.

Comparative Performance Analysis of Metaheuristic Feature Selection Methods for Speech Emotion Recognition

Abstract Emotion recognition systems from speech signals are realized with the help of acoustic or spectral features. Acoustic analysis is the extraction of digital features from speech files using digital signal processing methods. Another method is the analysis of time-frequency images of speech using image processing. The size of the features obtained by acoustic analysis is in the thousands. Therefore, classification complexity increases and causes variation in classification accuracy. In feature selection, features unrelated to emotions are extracted from the feature space and are expected to contribute to the classifier performance. Traditional feature selection methods are mostly based on statistical analysis. Another feature selection method is the use of metaheuristic algorithms to detect and remove irrelevant features from the feature set. In this study, we compare the performance of metaheuristic feature selection algorithms for speech emotion recognition. For this purpose, a comparative analysis was performed on four different datasets, eight metaheuristics and three different classifiers. The results of the analysis show that the classification accuracy increases when the feature size is reduced. For all datasets, the highest accuracy was achieved with the support vector machine. The highest accuracy for the EMO-DB, EMOVA, eNTERFACE’05 and SAVEE datasets is 88.1%, 73.8%, 73.3% and 75.7%, respectively.

Automatic detection of nasal closure and nasal release landmark acoustic cues

This study describes the detection of nasal closure and nasal release landmarks, as part of a larger system for speech recognition based on acoustic cues. Landmarks are produced as a result of closures and releases in the oral region and are indicated by abrupt changes in the speech signal. Nasal closure and release landmarks have proven particularly challenging to detect and are the focus of this report. The process for implementing the nasal detection module includes extracting and processing a set of speech-related measurements, such as formant frequencies, spectral band energies, and their derivatives, from a large database of labeled speech files, and determining which of these measurements are potentially effective, using ANOVA analysis. Next, Gaussian mixture models are trained and tested on these measurements to classify nasal closures, nasal releases, and all other landmark cues. The resulting nasal closure and release landmark detection module will be used with other landmark modules for vowels, glides, fricative closures/releases, and stop closures/releases, as well as other acoustic cues to place and voicing, in the overall speech recognition system. The current performance of the module will be assessed and discussed.

DSF Cryptography by Rotating Left the Samples Binary Values

A novel method of digital speech file cryptography will be proposed. The method will apply speech encryption-decryption at the bit level. The binary values of the speech samples will be shifted left to a defined number of rotating left digits. The private key will determine the length of the set of the bits to be rotated will be selected, the starting bit from where to start rotating and the number of rotating digits. The speech sample value will be represented in binary using 64_bits binary representation. Because of the fractional decimal value of the speech sample, changing the LSBs of the binary value may not affect the speech sample, so the starting bit for selecting the bits to be rotated will be investigated, and this bit will be used in the encryption-decryption process to get a damaged encrypted speech file. The private key will contains the starting bit; the bits set length and the number of rotating left digits, conditional selections of the private key value will be provided based on analyzing the quality of the encrypted speech files. The encryption and decryption process of the proposed method will be implemented in one round using a simple rotate left operation. The proposed method will not need key generation process; the values included in the private key will be directly used to apply speech encryption-decryption. The proposed method will be implemented using various speech files, the obtained results will be analyzed using quality, speed and sensitivity analysis to prove the efficiency of the proposed method.

Analysis of Covering Media Used to Hide DSF

Digital speech file is an important widely circulated type needed to be protected from being hacked. Data steganography is one of the easiest methods which can be used to protect the speech file. In this paper research the process of using each of the digital image or the digital speech as a covering media will be analyzes. LSB2 and LSB4 will be used for color images, while LSB32, LSB16, LSB8 and LSB4 will be used for a speech covering media, the quality of the obtained stego media will be analyzed to give the user the best choice of selecting the covering media foe digital speech hiding.

Quality Analysis of Using DSF and DCI as a Stego Medias

High quality of the stego media is an important issue to evaluate the stego system, the stego media must be closed to the covering one and the MSE measured between them must be low as possible, while the PSNR measured between them must be high as possible. In this paper research a detailed analysis of stego media will be produced, the quality of digital color image will be compared with the quality of digital speech file. Digital color image will be used to hold various messages, the same messages will be hidden in a digital speech file, and the quality parameters will be calculated. The results of using the image will be compared with the results of using digital speech file to raise some recommendation. 2 and 4 bits for message hiding will be tested using short, medium and long messages.

Production and Interpretation of Accentual Intonation Tunes in L2 Acquisition

Intonation is one of the most problematic aspects of the English prosody for L2 speakers. Studies on Nigerian English (NigE) intonation have claimed that Standard English intonation differs significantly from native Englishes but studies on the interpretation and production of accentual intonation tunes of L2 acquisition have been rare. Fifty final-year university undergraduates from Nigeria (25 males and 25 females) were purposively sampled. Each of the respondents was given 10 English sentences to produce and interpret. Interpretations of each of the respondents was captured and converted to simple percentages while the voice production of each of the respondents was elicited from a Speech Filing System (SFS) installed on Hp250 laptop. Respondents' frequency of occurrence at production level showed appropriate use of 134 (26.8%) while inappropriate use amounted to 366 (73.2%) out of 500 expected results of accentual tunes. Production on gender basis revealed males and females appropriate use at 67 (13.4%) each and inappropriate use was 183 (36.6%) respectively, both sexes showing no difference in accentual tunes use. Findings revealed that out of the 500 expected results for interpretation of accentual tunes, respondents appropriately interpreted the English sentences in 360 (72%) instances while 140 (28%) respondents misinterpreted the English sentences. Though, results of respondents revealed competence in the interpretation of accentual tunes (72%) but at the production level, respondents did not apply accentual tunes (26.8%) on English expressions. This study concludes that Nigerian spoken English accent differs significantly from native Englishes and should be described as an outer circle English.

Speech files cryptography using image key and indices key

Protecting speech files from being hacked is a vital issue. In this paper's research, a new and simple method of speech files cryptography will be introduced. The method will simplify the encryption and decryption functions by using a simple replacement operation. The method will provide a high level of security based on using two secret private keys: the image_key and the image position, these two keys will be used to generate an indices key, which will be used to apply speech sample replacements in the encryption and decryption phases. The introduced method will be efficiently used to encrypt-decrypt mono and stereo speeches. The method will be very flexible; the changes of the private keys will be an easy process, without the need to make any changes in the method algorithm. The proposed meth will be tested and implemented using various speech files, the obtained results will be used to prove the sensitivity quality, and efficiency of the proposed method.

Kencorpus: A Kenyan Language Corpus of Swahili, Dholuo and Luhya for Natural Language Processing Tasks

Indigenous African languages are categorized as under-served in Natural Language Processing. They therefore experience poor digital inclusivity and information access. The processing challenge with such languages has been how to use machine learning and deep learning models without the requisite data. The Kencorpus project intends to bridge this gap by collecting and storing text and speech data that is good enough for data-driven solutions in applications such as machine translation, question answering and transcription in multilingual communities. The Kencorpus dataset is a text and speech corpus for three languages predominantly spoken in Kenya: Swahili, Dholuo and Luhya (three dialects of Lumarachi, Lulogooli and Lubukusu). Data collection was done by researchers who were deployed to the various data collection sources such as communities, schools, media, and publishers. The Kencorpus' dataset has a collection of 5,594 items, being 4,442 texts of 5.6 million words and 1,152 speech files worth 177 hours. Based on this data, other datasets were also developed such as Part of Speech tagging sets for Dholuo and the Luhya dialects of 50,000 and 93,000 words tagged respectively. We developed 7,537 Question-Answer pairs from 1,445 Swahili texts and also created a text translation set of 13,400 sentences from Dholuo and Luhya into Swahili. The datasets are useful for downstream machine learning tasks such as model training and translation. Additionally, we developed two proof of concept systems: for Kiswahili speech-to-text and a machine learning system for Question Answering task. These proofs provided results of a performance of 18.87% word error rate for the former, and 80% Exact Match (EM) for the latter system. These initial results give great promise to the usability of Kencorpus to the machine learning community. Kencorpus is one of few public domain corpora for these three low resource languages and forms a basis of learning and sharing experiences for similar works especially for low resource languages. Challenges in developing the corpus included deficiencies in the data sources, data cleaning challenges, relatively short project timelines and the Coronavirus disease (COVID-19) pandemic that restricted movement and hence the ability to get the data in a timely manner.

Pausing before verb production is associated with mild cognitive impairment in Parkinson's disease.

Cognitive dysfunction and communication impairment are common and disabling symptoms in Parkinson's Disease (PD). Action verb deficits occur in PD, but it remains unclear if these impairments are related to motor system dysfunction and/or cognitive decline. The objective of our study was to evaluate relative contributions of cognitive and motor dysfunction to action verb production in naturalistic speech of patients with PD. We proposed that pausing before action-related language is associated with cognitive dysfunction and may serve as a marker of mild cognitive impairment in PD. Participants with PD (n = 92) were asked to describe the Cookie Theft picture. Speech files were transcribed, segmented into utterances, and verbs classified as action or non-action (auxiliary). We measured silent pauses before verbs and before utterances containing verbs of different classes. Cognitive assessment included Montreal Cognitive Assessment (MoCA) and neuropsychological tests to categorize PD participants as normal cognition (PD-NC) or mild cognitive impairment (PD-MCI) based on Movement Disorders Society (MDS) Task Force Tier II criteria. Motor symptoms were assessed using MDS-UPDRS. We performed Wilcoxon rank sum tests to identify differences in pausing between PD-NC and PD-MCI. Logistic regression models using PD-MCI as dependent variables were used to evaluate the association between pause variables and cognitive status. Participants with PD-MCI demonstrated more pausing before and within utterances compared to PD-NC, and the duration of these pauses were correlated with MoCA but not motor severity (MDS-UPDRS). Logistic regression models demonstrated that pauses before action utterances were associated with PD-MCI status, whereas pauses before non-action utterances were not significantly associated with cognitive diagnosis. We characterized pausing patterns in spontaneous speech in PD-MCI, including analysis of pause location with respect to verb class. We identified associations between cognitive status and pausing before utterances containing action verbs. Evaluation of verb-related pauses may be developed into a potentially powerful speech marker tool to detect early cognitive decline in PD and better understand linguistic dysfunction in PD.

Speech Files Cryptography using Indices Keys and Blocking

Protecting speech file is a vital issue. In this paper research a simple method of speech cryptography will be introduced. The method will use indices key to perform simple replacement operations to apply speech encryption and decryption, the replacement operation will use the indexes and the indexes contents to apply speech cryptography. The proposed method will protect the speech signal by increasing the security level, the method will use a color image as an image_key, this key will be kept in secret between the sender and receiver and it will be used to generate the indices key. The proposed will sensitive to the selected image_key, changing this image will be considered as a hacking attempt by producing a damaged decrypted speech signal. The proposed method will be flexible, changing the image_key will not cause any changes in the proposed method algorithms. To increase the efficiency of the proposed method by decreasing the processing time the speech file will be divided into blocks, and each block will be treated separately. The proposed method will be tested and implemented using various speech files, the obtained results will be used to prove the quality, sensitivity and efficiency of the proposed method.

Broad phonetic feature classifier for real-time hearing aid processing

Current Hearing Aid (HA) devices compensate for hearing loss by amplifying sounds based on instantaneous energy estimates in specific frequency bands. The dynamics of speech are incorporated using attack and release times—a crude approximation of the rich temporal dynamics in natural speech. Recent advances in silicon enables one to consider more complex sound processing approaches with minimal impact on the HA battery life. We envisage a broad phonetic feature classifier, followed by amplification strategies optimized for specific temporal dynamics across each segment of natural speech. In this work we present a broad phonetic classifier for ‘vocalic’, ‘non-vocalic’ and ‘mixed’ speech segments. The classifier is based on (i) Interpretable models- unsupervised learning methods such as Gaussian Mixture Models followed by supervised learning based Radial Basis Function networks; and (ii) Deep Neural Networks with supervised learning. The frontend feature extractor consists of MFCC, delta MFCC features, spectral flatness measure and segmental energy. The training and test sets are generated using speech files and the corresponding phone labels from the TIMIT dataset. We present our findings with respect to the classifier's performance and deployability for real time HA processing.

Performance Comparison between RSA and El-Gamal Algorithms for Speech Data Encryption and Decryption

This article presents a performance comparison of two known public key cryptography techniques namely RSA (Rivest–Shamir–Adleman) and El-Gamal algorithms to encrypt/decrypt the speech signals during transferring over open networks. Specifically, this work is divided into two stages. The first stage is enciphering-deciphering the input speech file by employing the RSA method. The second stage is enciphering-deciphering the same input speech file by employing the El-Gamal method. Then, a comparative analysis is performed to test the performance of both cryptosystems using diverse experimental and statistical analyses for the ciphering and deciphering procedures like some known speech quality measures: histogram, spectrogram, correlation, differential, speed performance and noise effect analyses. The analyses outcomes reveal that the RSA and El-Gamal approaches are efficient and adequate for providing high degree of security, confidentiality and reliability. Additionally, the outcomes indicate that the RSA speech cryptosystem outperforms its counterpart the El-Gamal speech cryptosystem in most of ciphering/deciphering speech performance metrics.

A Pitch Estimation Algorithm for Speech in Complex Noise Environments Based on the Radon Transform

The pitch period as an essential feature is used in various speech-related works. Most actual projects collect speech signals in complex noise environments. Thus, the noise resistance of the algorithm for accurate pitch estimation has become more critical than ever. However, many state-of-the-art algorithms fail to obtain good results when dealing with noisy speech files at a low signal-to-noise ratio (SNR) value. This study presents a new noise-resistant pitch estimation algorithm based on the Radon transform and reduces the influence of formants with the modification of the classical equation. In addition, we use the difference between the pitch candidates of the consecutive frames as part of the criterion for the decoding of the Viterbi algorithm to strengthen the correlation of the pitch estimates and make the pitch contours smoother. We synthesized three noisy speech databases with 18 types of collected environmental noise and compared our algorithm with 7 state-of-the-art algorithms. The proposed algorithm has the best performance on CSTR and self-recorded databases and reduces Gross Pitch Error (GPE) rate by over 12% at 0 dB SNR against Bayesian Pitch Tracker. In particular, the GPE rate of our proposed algorithm can be maintained under 25% at 0 dB SNR, while BaNa only achieves 35%.

Robust intelligibility and quality evaluation of combined temporal and spectral processing for hearing impaired

• Combined temporal-spectral processing objectively evaluated in terms of HASQI and HASPI. • Six types of audiograms based on different slope and frequency losses are selected. • -5, -3, 0, 3, 5, 10 dB SNRs noisy environments in cafe, traffic, station and train are considered. • Proposed enhanced algorithm shows markable improvements in quality and intelligibility. Hearing-impaired people face numerous challenges with speech perception in the presence of interfering background noise. To suppress interfering background noise, the common approach widely used is speech enhancement. Inspired by the improved results of combined temporal and spectral processing in speech enhancement, this research study proposes temporal enhancement combined with two different spectral enhancement methods, with a novel approach of soft masking using priori and posterior signal to noise ratio uncertainty. The present study investigates quality and intelligibility objective evaluations, namely, hearing aid speech quality index and hearing aid speech perception index, of spectral and a combination of temporal-spectral speech enhancement methods for typical pattern of hearing loss characterized by six audiograms. For evaluation, clean speech files from the NOIZEUS database are mixed with four local noises, namely, cafeteria, traffic, station, and train at -5, -3, 0,3,5, and 10 dB SNRs. These local noises are quite common, which are encountered by people in their day-to-day lives. In most of the testing conditions, the new combined temporal spectral enhancement shows improved results in comparison with the purely spectral processing methods.

Compressing English Speech Data with Hybrid Methods without Data Loss

Understanding the mechanism of speech formation is of great importance in the successful coding of the speech signal. It is also used for various applications, from authenticating audio files to connecting speech recording to data acquisition device (e.g. microphone). Speech coding is of vital importance in the acquisition, analysis and evaluation of sound, and in the investigation of criminal events in forensics. For the collection, processing, analysis, extraction and evaluation of speech or sounds recorded as audio files, which play an important role in crime detection, it is necessary to compress the audio without data loss. Since there are many voice changing software available today, the number of recorded speech files and their correct interpretation play an important role in detecting originality. Using various techniques such as signal processing, noise extraction, filtering on an incomprehensible speech recording, improving the speech, making them comprehensible, determining whether there is any manipulation on the speech recording, understanding whether it is original, whether various methods of addition and subtraction are used, coding of sounds, the code must be decoded and the decoded sounds must be transcribed. In this study, first of all, what sound coding is, its purposes, areas of use, classification of sound coding according to some features and techniques are given. Moreover, in our study speech coding was done on the English audio data. This dataset is the real dataset and consists of approximately 100000 voice recordings. Speech coding was done using waveform, vocoders and hybrid methods and the success of all the methods used on the system we created was measured. Hybrid models gave more successful results than others. The results obtained will set an example for our future work.