Rahul Kumar Jaiswal

• Jaiswal, Rahul Kumar; Elnourani, Mohamed; Deshmukh, Siddharth & Beferull-Lozano, Baltasar (2025). A Data-driven Transfer Learning Method for Indoor Radio Map Estimation. IEEE Transactions on Vehicular Technology. ISSN 0018-9545. 75(3), s. 4261–4277. doi: 10.1109/tvt.2025.3609207. Fulltekst i vitenarkiv Vis sammendrag

  Estimating accurate radio maps is important for various tasks in wireless communications, such as localization, resource allocation, and network planning. Due to the changes in the propagation characteristics of the wireless environments, a radio map model learned under a particular wireless environment cannot be directly used in a new wireless environment. Moreover, learning a new model for every environment requires, in general, a large amount of data and is computationally demanding. In this work, we design an effective novel data-driven transfer learning method that transfers and fine-tunes a deep neural network (DNN)-based radio map model learned from an original indoor wireless environment to other indoor environments with a certain level of similarity. Since other widely used similarity measures do not consider the wireless propagation characteristics, we design a data-driven similarity measure that predicts the mean square error (MSE) of the estimated power values and the amount of training data needed when learning a radio map in a new environment. Extensive simulations over several wireless environments of both office and cafe area environments show that the proposed method achieves savings of approximately 40-90% in sensor measurement data while maintaining similar accuracy (MSE) as a model trained without transfer learning, and outperforms state-of-the-art methods.

• Jaiswal, Rahul Kumar; Elnourani, Mohamed; Deshmukh, Siddharth & Beferull-Lozano, Baltasar (2025). Leveraging Transfer Learning for Radio Map Estimation via Mixture of Experts. IEEE Transactions on Cognitive Communications and Networking (TCCN). ISSN 2372-2045. 12, s. 846–863. doi: 10.1109/TCCN.2025.3570469. Fulltekst i vitenarkiv Vis sammendrag

  This paper leverages transfer learning (TL) on a mixture of experts (MoE) model for indoor radio map estimation. The proposed MoE combines location-based and location-free experts through a gating network exploiting their complementary benefits. To estimate the radio map in a new wireless environment, the learned model of another sufficiently similar wireless environment is transferred and fine-tuned with additional data from the new wireless environment. The proposed data-driven similarity measure predicts the amount of training data needed for TL. Results demonstrate that the proposed method achieves similar accuracy, in terms of mean square error (MSE), to a model trained without TL while only requiring 5-40% of measurement data to adapt to several varying wireless environments. The test environments include both the office area and the cafe area wireless environments. As expected, the proposed MoE method outperforms both experts as well as the state-of-the-art methods, in both the presence and the absence of noise in location-based and location-free features.

• Jaiswal, Rahul Kumar; Elnourani, Mohamed; Deshmukh, Siddharth & Beferull-Lozano, Baltasar (2023). Location-free Indoor Radio Map Estimation using Transfer learning, 2023 IEEE 97th Vehicular Technology Conference (VTC2023-Spring). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9798350311143. doi: 10.1109/VTC2023-Spring57618.2023.10200979. Fulltekst i vitenarkiv Vis sammendrag

  Accurate estimation of radio maps is important for various applications of wireless communications, such as network planning, and resource allocation. To learn accurate radio map models, one needs to have accurate knowledge of transmitter and receiver locations. However, it is difficult to obtain accurate locations in practice, especially, in scenarios having a high degree of wireless multi-path. Alternatively, time of arrival (ToA) features, which are easier to obtain, can be employed for estimating radio maps. To this end, this paper investigates the application of transfer learning method using ToA features for estimating radio maps under indoor wireless communications. The performance is compared with the scenarios where only the locations of receivers and both ToAs and locations of receivers, are used for estimating radio maps, assuming that locations are known. Due to the changes in propagation characteristics, a radio map model learned in a specific wireless environment cannot be directly employed in a new wireless environment. To address this issue, a data-driven transfer learning method is designed that transfers and fine-tunes a deep neural network model learned for a radio map from a source wireless environment to other distinct (target) wireless environments. Our proposed method predicts the training data required in the new wireless environments using a data-driven similarity measure. Our results demonstrate that using ToA (location-free) features results in a superior performance for estimating radio maps in terms of the necessary number of sensor measurements for estimating radio maps with a good accuracy, as compared to a location-based approach, where it may be difficult to have accurate location estimations. It leads to a saving of 70-90% of the necessary sensor measurement data for a mean square error (MSE) of 0.004.

• Jaiswal, Rahul Kumar & Dubey, Rajesh Kumar (2023). Multiple time-instances features based approach for reference-free speech quality measurement. Computer Speech and Language. ISSN 0885-2308. 79. doi: 10.1016/j.csl.2022.101478. Fulltekst i vitenarkiv Vis sammendrag

  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%.

• Jaiswal, Rahul Kumar & Dubey, Rajesh Kumar (2023). CAQoE: A Novel No-Reference Context-aware Speech Quality Prediction Metric. ACM Transactions on Multimedia Computing, Communications, and Applications (TOMCCAP). ISSN 1551-6857. 19(1), s. 1–23. doi: 10.1145/3529394. Fulltekst i vitenarkiv Vis sammendrag

  The quality of speech degrades while communicating over Voice over Internet Protocol applications, for example, Google Meet, Microsoft Skype, and Apple FaceTime, due to different types of background noise present in the surroundings. It reduces human perceived Quality of Experience (QoE). Along this line, this article proposes a novel speech quality prediction metric that can meet human’s desired QoE level. Our motivation is driven by the lack of evidence showing speech quality metrics that can distinguish different noise degradations before predicting the quality of speech. The quality of speech in noisy environments is improved by speech enhancement algorithms, and for measuring and monitoring the quality of speech, objective speech quality metrics are used. With the integration of these components, a novel no-reference context-aware QoE prediction metric (CAQoE) is proposed in this article, which initially identifies the context or noise type or degradation type of the input noisy speech signal and then predicts context-specific speech quality for that input speech signal. It will have of great importance in deciding the speech enhancement algorithms if the types of degradations causing poor speech quality are known along with the quality metric. Results demonstrate that the proposed CAQoE metric outperforms in different contexts as compared to the metric where contexts are not identified before predicting the quality of speech, even in the presence of limited size speech corpus having different contexts available from the NOIZEUS speech database.

• Jaiswal, Rahul Kumar; Deshmukh, Siddharth; Elnourani, Mohamed & Beferull-Lozano, Baltasar (2022). Transfer Learning Based Joint Resource Allocation for Underlay D2D Communications, 2022 IEEE Wireless Communications and Networking Conference (WCNC). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781665442664. s. 1479–1484. doi: 10.1109/WCNC51071.2022.9771636. Fulltekst i vitenarkiv Vis sammendrag

  In this paper, we investigate the application of transfer learning to train a Deep Neural Network (DNN) model for joint channel and power allocation in underlay device-todevice (D2D) communication. Based on the traditional optimization solutions, generating training dataset for scenarios with perfect channel state information (CSI) is not computationally demanding, compared to scenarios with imperfect CSI. Thus, a transfer learning-based approach can be exploited to transfer the DNN model trained for the perfect CSI scenarios to the imperfect CSI scenarios. We also consider the issue of defining the similarity between two types of resource allocation tasks. For this, we first determine the value of outage probability for which two resource allocation tasks are same, that is, for which our numerical results illustrate the minimal need of relearning from the transferred DNN model. For other values of outage probability, there is a mismatch between the two tasks and our results illustrate a more efficient relearning of the transferred DNN model. Our results show that the learning dataset required for relearning of the transferred DNN model is significantly smaller than the required training dataset for a DNN model without transfer learning.

• Jaiswal, Rahul Kumar; Elnourani, Mohamed; Deshmukh, Siddharth & Beferull-Lozano, Baltasar (2022). Deep Transfer Learning Based Radio Map Estimation for Indoor Wireless Communications, 2022 IEEE 23rd International Workshop on Signal Processing Advances in Wireless Communication (SPAWC). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781665494557. doi: 10.1109/SPAWC51304.2022.9833974. Fulltekst i vitenarkiv Vis sammendrag

  This paper investigates the problem of transfer learning in radio map estimation for indoor wireless communications, which can be exploited for different applications, such as channel modelling, resource allocation, network planning, and reducing the number of necessary power measurements. Due to the nature of wireless communications, a radio map model developed under a particular environment can not be directly used in a new environment because of the changes in the propagation characteristics, thus creating a new model for every environment requires in general a large amount of data and is computationally demanding. To address these issues, we design an effective novel data-driven transfer learning procedure that transfers and fine-tunes a deep neural network (DNN)-based model for a radio map learned from an original indoor wireless environment to other different indoor wireless environments. Our method allows to predict the amount of training data needed in new indoor wireless environments when performing the operation of transfer learning using our similarity measure. Our simulation results illustrate that the proposed method achieves a saving of 60-70% in sensor measurement data and is able to adapt to a new wireless environment with a small amount of additional data.

• Jaiswal, Rahul Kumar (2022). Automatic Noise Class Detection For Improving Speech Quality Using Artificial Neural Networks, 2022 7th International Conference on Communication and Electronics Systems (ICCES). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781665496346. s. 1285–1288. doi: 10.1109/ICCES54183.2022.9836006. Fulltekst i vitenarkiv Vis sammendrag

  Speech quality degrades in noisy environments while using a particular voice over internet protocol (VoIP) application, for example, Microsoft Skype, Apple Facetime, etc. Speech enhancement algorithms separate noise from the degraded speech. No-reference speech quality metric (SQM), such as P.563, measures the quality of speech. To this end, this research study develops a novel approach for extracting features from the noisy speech samples available from the NOIZEUS corpus for detecting the noise class (noise type and SNR) using deep neural network (DNN). It integrates speech enhancement algorithms with SQM to estimate the speech quality (MOS score) from the noisy samples, which is then used as a feature vector to train a DNN. Results demonstrate that the DNN outperforms in detecting the noise class as compared to the machine learning classifiers, tested with different noise classes. This suggests to develop a noise-sensitive speech quality prediction model for real-time measuring and monitoring of the quality of speech.

• Jaiswal, Rahul Kumar; Yeduri, Sreenivasa Reddy & Cenkeramaddi, Linga Reddy (2022). Single-channel speech enhancement using implicit Wiener filter for high-quality speech communication. International Journal of Speech Technology. ISSN 1381-2416. 25(3), s. 745–758. doi: 10.1007/s10772-022-09987-4. Fulltekst i vitenarkiv Vis sammendrag

  Speech enables easy human-to-human communication as well as human-to-machine interaction. However, the quality of speech degrades due to background noise in the environment, such as drone noise embedded in speech during search and rescue operations. Similarly, helicopter noise, airplane noise, and station noise reduce the quality of speech. Speech enhancement algorithms reduce background noise, resulting in a crystal clear and noise-free conversation. For many applications, it is also necessary to process these noisy speech signals at the edge node level. Thus, we propose implicit Wiener filter-based algorithm for speech enhancement using edge computing system. In the proposed algorithm, a first order recursive equation is used to estimate the noise. The performance of the proposed algorithm is evaluated for two speech utterances, one uttered by a male speaker and the other by a female speaker. Both utterances are degraded by different types of non-stationary noises such as exhibition, station, drone, helicopter, airplane, and white Gaussian stationary noise with different signal-to-noise ratios. Further, we compare the performance of the proposed speech enhancement algorithm with the conventional spectral subtraction algorithm. Performance evaluations using objective speech quality measures demonstrate that the proposed speech enhancement algorithm outperforms the spectral subtraction algorithm in estimating the clean speech from the noisy speech. Finally, we implement the proposed speech enhancement algorithm, in addition to the spectral subtraction algorithm, on the Raspberry Pi 4 Model B, which is a low power edge computing device.

• Dayal, Aveen; Yeduri, Sreenivasa Reddy; Koduru, Balu Harshavardan; Jaiswal, Rahul Kumar; J, Soumya & B., Srinivas M. [Vis alle 8 forfattere av denne artikkelen] (2022). Lightweight deep convolutional neural network for background sound classification in speech signals. Journal of the Acoustical Society of America. ISSN 0001-4966. 151(4), s. 2773–2786. doi: 10.1121/10.0010257. Fulltekst i vitenarkiv Vis sammendrag

  Recognizing background information in human speech signals is a task that is extremely useful in a wide range of practical applications, and many articles on background sound classification have been published. It has not, however, been addressed with background embedded in real-world human speech signals. Thus, this work proposes a lightweight deep convolutional neural network (CNN) in conjunction with spectrograms for an efficient background sound classification with practical human speech signals. The proposed model classifies 11 different background sounds such as airplane, airport, babble, car, drone, exhibition, helicopter, restaurant, station, street, and train sounds embedded in human speech signals. The proposed deep CNN model consists of four convolution layers, four max-pooling layers, and one fully connected layer. The model is tested on human speech signals with varying signal-to-noise ratios (SNRs). Based on the results, the proposed deep CNN model utilizing spectrograms achieves an overall background sound classification accuracy of 95.2% using the human speech signals with a wide range of SNRs. It is also observed that the proposed model outperforms the benchmark models in terms of both accuracy and inference time when evaluated on edge computing devices.

• Jaiswal, Rahul Kumar & Dubey, Rajesh Kumar (2022). Non-intrusive speech quality assessment using context-aware neural networks. International Journal of Speech Technology. ISSN 1381-2416. 25. doi: 10.1007/s10772-022-10011-y. Fulltekst i vitenarkiv Vis sammendrag

  To meet the human perceived quality of experience (QoE) while communicating over various Voice over Internet protocol (VoIP) applications, for example Google Meet, Microsoft Skype, Apple FaceTime, etc. a precise speech quality assessment metric is needed. The metric should be able to detect and segregate different types of noise degradations present in the surroundings before measuring and monitoring the quality of speech in real-time. Our research is motivated by the lack of clear evidence presenting speech quality metric that can firstly distinguish different types of noise degradations before providing speech quality prediction decision. To that end, this paper presents a novel non-intrusive speech quality assessment metric using context-aware neural networks in which the noise class (context) of the degraded or noisy speech signal is first identified using a classifier then deep neutral networks (DNNs) based speech quality metrics (SQMs) are trained and optimized for each noise class to obtain the noise class-specific (context-specific) optimized speech quality predictions (MOS scores). The noisy speech signals, that is, clean speech signals degraded by different types of background noises are taken from the NOIZEUS speech corpus. Results demonstrate that even in the presence of less number of speech samples available from the NOIZEUS speech corpus, the proposed metric outperforms in different contexts compared to the metric where the contexts are not classified before speech quality prediction.

• Jaiswal, Rahul Kumar (2021). Influence of Silence and Noise Filtering on Speech Quality Monitoring, 2021 International Conference on Speech Technology and Human-Computer Dialogue (SpeD). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781665427869. s. 109–113. doi: 10.1109/SpeD53181.2021.9587364. Fulltekst i vitenarkiv Vis sammendrag

  With the exponential increase of mobile users and internet subscribers, the utilization of voice over internet protocol (VoIP) application is increasing dramatically. People exploit different VoIP applications for effective communication, for example, Google Meet, Microsoft Skype, Zoom video conferencing applications, etc. The single-ended speech quality metrics are employed for measuring and monitoring the quality of speech. However, different types of degradations present in the surroundings distort the quality of speech. In order to meet the desired quality of experience (QoE) level of end-user while using VoIP applications, it is necessary to reduce VoIP degradations and obtain the optimized speech quality. Along that line, this paper investigates the conjunction of filtering of silence and noise as a pre-processing block with the single-ended speech quality metric under various common occurring degradations encountered during VoIP communication. This can help the internet service providers in understanding the potential root cause of decrement in quality of speech and then applying the QoE management service to fulfill desired human QoE level. Results demonstrate that the deployment of joint pre-processing on speech samples under various VoIP degradations improves the quality of speech to a great extent.

• Jaiswal, Rahul Kumar & Dubey, Rajesh Kumar (2021). Concatenative Text-to-Speech Synthesis System for Communication Recognition, 2021 5th International Conference on Electronics, Communication and Aerospace Technology (ICECA). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781665435246. s. 867–872. doi: 10.1109/ICECA52323.2021.9675855. Fulltekst i vitenarkiv Vis sammendrag

  Text-to-speech (TTS) synthesis is one of the rapidly emerging areas of computer-to-human interaction technology. Human-like speech is replicated by the computer with the introduction of input text which is usually very natural. Real-life applications of TTS synthesis technique make users task hassle-free. For example, reading book for the visually impaired people, paying electricity bill through automated call-centre, announcing train information at the railway station, etc. Traditionally, rule-based speech synthesis methods are deployed which find difficulties in obtaining optimal rules, resulting in lack of naturalness in the generated synthesized speech. Alternatively, to meet the desired quality of experience (QoE) of users while using these applications, this paper designs and develops a simple and robust TTS synthesis system for English language using the concatenative speech synthesis method and its variants and finds its suitability in intelligible and/or natural speech production. Various steps involved in processing text for speech production through TTS system are described. Results demonstrate that the speech generated by the concatenative speech synthesis method, in particular, unit-selection technique is smoother and natural, sounding like human voice. This is supported by informal listening test of the generated synthesized speech.

• Jaiswal, Rahul Kumar & Romero, Daniel (2021). Implicit Wiener Filtering for Speech Enhancement In Non-Stationary Noise, 2021 11th International Conference on Information Science and Technology (ICIST). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781665412667. s. 39–47. doi: 10.1109/ICIST52614.2021.9440639. Fulltekst i vitenarkiv Vis sammendrag

  Speech quality is degraded in the presence of background noise, which reduces the quality of experience (QoE) of the end-user and therefore motivates the usage of speech enhancement algorithms. A large number of approaches have been proposed in this context. However most of them have focused on the case where the noise is stationary, an assumption that seldom holds in practice. For instance, in mobile telephony, noise sources with a marked non-stationary spectral signature include vehicles, machines, and other speakers to name a few. On the other hand, the usage of frequency-domain information in existing algorithms for speech enhancement in non-stationary noise environments can be made more effective by leveraging the increased flexibility introduced by implicit Wiener filters, which allow the control of the spectral reconstruction of the speech signal through the adjustment of hyperparameters. To address these limitations, the present paper develops an algorithm that recursively estimates the noise power spectral density and reconstructs the target speech signal in the frequency domain by means of an implicit Wiener filter with judiciously selected hyperparameters. The recursive noise estimation approach relies on the past and the present power spectral values. To evaluate the performance of the speech enhancement algorithm, speech uttered by a male and a female speaker degraded by non-stationary noise produced e.g. by babbling, cars, street noise, trains, restaurants, and airport noise. To this end, the NOIZEUS corpus is used. Objective speech quality measures such as the log-likelihood ratio (LLR), the cepstral distance (CD), and the weighted spectral slope distance (WSS) are evaluated for the enhanced speech signals and compared to the conventional spectral subtraction method. Results demonstrate that the proposed algorithm provides consistent and improved enhancement performance with all tested noise types.

• Jaiswal, Rahul Kumar & Hines, Andrew (2020). Towards a Non-Intrusive Context-Aware Speech Quality Model, 2020 31st Irish Signals and Systems Conference (ISSC). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781728194189. s. 1–5. doi: 10.1109/ISSC49989.2020.9180171. Fulltekst i vitenarkiv Vis sammendrag

  Understanding how humans judge perceived speech quality while interacting through Voice over Internet Protocol (VoIP) applications in real-time is essential to build a robust and accurate speech quality prediction model. Speech quality is degraded in the presence of background noise reducing the Quality of Experience (QoE). Speech Enhancement (SE) algorithms can improve speech quality in noisy environments. The publicly available NOIZEUS speech corpus contains speech in environmental background noise babble, car, street, and train at two Signal-to-noise ratio (SNRs) 5dB and 10dB. Objective Speech Quality Metrics (OSQM) are used to monitor and measure speech quality for VoIP applications. This paper proposes a Context-aware QoE prediction model, CAQoE, which classifies the speech signal context (i.e., noise type and SNR) in order to allow context-specific speech quality prediction. This paper presents experiments conducted to develop the speech context-classification component of the proposed CAQoE model. Speech enhancement algorithms are used in conjunction with an OSQM to estimate Mean Opinion Score (MOS) of noisy and enhanced samples in order to train Machine Learning (ML) classifiers to classify the speech signal context (i.e., noise type and SNR). Results demonstrate that a Decision Tree (DT) classifier has better classification accuracy for the noise classes tested. We present the associated components of the CAQoE model, namely; Voice Activity Detection (VAD) and Speech Quality Model (SQM).

Se alle arbeider i NVA

• Jaiswal, Rahul Kumar (2025). Leveraging Machine Learning for Fraud Identification in International Shipping. doi: 10.1109/icsc64553.2025.10968567. Fulltekst i vitenarkiv Vis sammendrag

  With a surge in global trade between several countries, there has been a high rise in fraud cases involving the wrong declaration of goods by domestic parties. Traditionally, random inspections are done by customs officers, and the finding of fraud is often by chance. To alleviate this challenge, an intelligent data-driven machine learning (ML) model must identify frauds efficiently. In this paper, we design and develop a simple ML model to identify frauds effectively. We use shipment data comprising several features used during the custom declaration. We investigate the fraud identification for different possible feature combinations. Moreover, we also compare our model with different ML models. Results demonstrate that with all features combined, the ML model based on the Isolation Forest outperforms. This signifies that fraudsters can be easily identified through customs officers' inspection of domestic parties.

• Jaiswal, Rahul (2023). Deep Learning Based Speech Quality Assessment Focusing on Noise Effects. doi: 10.1007/978-3-031-48312-7_22. Fulltekst i vitenarkiv Vis sammendrag

  This paper investigates the suitability of different speech features in measuring and monitoring speech quality in order to fulfil the expected level of human perceived quality of experience (QoE) while using applications, such as Microsoft Skype, and Apple FaceTime to name a few. To this end, two speech features, namely; line spectral frequencies (LSF), and multi-resolution auditory model (MRAM) are extracted from the speech signal after processing it through a voice activity detector (VAD). A series of deep neural network (DNN)-based objective no-reference speech quality models (SQMs) are then developed employing a single speech feature and combining both speech features. Two noisy speech datasets, namely; Supplement-23 and NOIZEUS-2240 are used for the experiment. Simulation results demonstrate that the SQM developed using combined speech features results in a better speech quality prediction as compared to the SQM developed using a single speech feature, when tested with distinct types of speech degradations.

• Jaiswal, Rahul Kumar; Elnourani, Mohamed; Deshmukh, Siddharth & Beferull-Lozano, Baltasar (2023). Location-free Indoor Radio Map Estimation using Transfer learning. Fulltekst i vitenarkiv
• Jaiswal, Rahul Kumar (2022). Performance Analysis of Voice Activity Detector in Presence of Non-stationary Noise. doi: 10.1007/978-981-16-8129-5_10. Fulltekst i vitenarkiv Vis sammendrag

  Speech is degraded in the presence of background noise. The need to detect the presence of voiced segments accurately in the degraded signal is crucial for many speech processing applications. This paper addresses the problem of separation of speech and non-speech (noise/silence) segments under non-stationary noisy environments by means of Voice Activity Detector (VAD). A VAD detects the speech and non-speech segments by extracting the speech features and comparing it to a threshold. In this paper, the VAD algorithms are based on two speech features: energy and spectral centroid. NOIZEUS speech corpus containing speech degraded by non-stationary noises at four different SNRs are used. The performance of the VAD algorithms is evaluated using F-score and Euclidean distance with comparison to the Ground truth VAD. Results demonstrate that for different noise conditions tested, a weighted spectral centroid VAD achieves outstanding performance.

• Jaiswal, Rahul Kumar (2022). Performance Analysis of Deep Learning Based Speech Quality Model with Mixture of Features. doi: 10.1109/ism55400.2022.00053. Fulltekst i vitenarkiv Vis sammendrag

  Speech is one of the convenient mediums of communication among humans. However, the quality of speech deteriorates due to the surrounding noise. To fulfil the expected level of quality of experience (QoE) of the end-user while exploiting distinct applications, such as Microsoft Skype, Apple FaceTime to name a few, it is important to measure the speech quality and monitor it in the real-time. To this end, this paper investigates a series of deep neural network (DNN)-based objective no-reference speech quality models (SQMs) in accurately measuring speech quality. Three speech features, namely, line spectral frequencies (LSF), mel-frequency cepstral coefficients (MFCC), and multi-resolution auditory model (MRAM) are extracted from the speech signal after processing it through a voice activity detector (VAD). A series of DNN-based SQMs is, then, developed by incorporating either a single or a mixture of speech features. The standard no-reference speech quality prediction model (P.563) is employed as a baseline model. Results demonstrate that the DNN-based SQM trained with MRAM feature performs better in accurately measuring speech quality as compared to the baseline model and other DNN-based SQMs trained with different speech features or their mixtures.

• Jaiswal, Rahul Kumar (2021). Speech Activity Detection under Adverse Noisy Conditions at Low SNRs. doi: https:/ieeexplore.ieee.org/abstract/document/9488934. Fulltekst i vitenarkiv Vis sammendrag

  Speech originating from the noisy environments degrades the speech quality and intelligibility, thus reducing the human perceived Quality of Experience (QoE). For example, surveillance using drone during natural catastrophe needs an efficient speech recognition device to recognise the speech of the frozen human in presence of drone noise to save their life. Therefore, it often requires to pre-process the noisy speech in order to reduce the noise artifacts and enhance the speech. This paper detects the speech activity using Voice Activity Detection (VAD). The VAD distinguishes speech activity (speech presence) and speech inactivity (silence/noise) by extracting the speech features and comparing to a threshold. The energy and spectral centroid features are deployed to design VADs. Noisy dataset consisting of urban noise, for example, drone, helicopter, airplane and station noise, is created at different signal-to-noise ratios (SNRs). F-score and Euclidean distance are used to measure the performance of VADs. Results demonstrate that the spectral centroid VAD performs outstanding with various noise degradations tested.

• Jaiswal, Rahul Kumar (2024). Data-driven Transfer Learning Methods for Wireless Networks. Universitetet i Agder. ISSN 9788284271927. Fulltekst i vitenarkiv Vis sammendrag

  Radio maps provide information about spatial signal strength and network coverage in a designated geographical area. The estimation of accurate radio maps is necessary to improve the performance of many applications of future wireless networks. For instance, localization, network planning, and resource allocation, to name a few. To obtain accurate radio maps, the exact knowledge of transmitter(Tx) and receiver(Rx) locations can be used. This is known as the location-based method. However, in practice, wireless networks incur a high degree of multipath. As a result, it is difficult to obtain accurate locations of Rxs. Alternatively, time of arrival (ToA) features of radio signals, which are easier to obtain, can be used. This is known as the location-free method. One of the ways to incorporate both methods is the mixture of experts (MoE). Due to changes in the propagation characteristics of wireless networks, a radio map model designed under a particular wireless environment (source environment) can not be directly used in a new wireless environment (target environment). Moreover, designing a new radio map model for each new wireless environment requires a huge amount of measurement samples and may need substantial computational resources and data acquisition costs. To address these issues, in this dissertation, we propose a series of transfer learning (TL) schemes using each of the aforementioned methods to estimate radio maps in new wireless environments where there is a scarcity of measurement samples. To this end, we first train a radio map model in a source wireless environment and then transfer it to another similar but still different target wireless environment. It is then fine-tuned using a small amount of samples of the target wireless environment to reduce the data acquisition cost. For such a scheme, the similarity between two wireless environments controls the effectiveness of the TL operation. Therefore, to quantify the similarity, we investigate different classical similarity measures including the widely used Wasserstein distance. Numerically, we show that these classical measures do not perform well in the context of TL for radio map estimation. To overcome the limitations of these classical measures, we design a data-driven similarity measure (DDS), which is able to capture all the variations of wireless environments and can learn the wireless propagation characteristics directly from the data. Additionally, our DDS can predict the amount of training data needed to estimate radio maps in new target wireless environments when performing the TL operation. Experiments show that our proposed TL schemes perform efficiently with high model accuracy and save a substantial amount of sensor measurement data. Different models are designed for each of the cases of location-based, location-free, and MoE-based radio map estimation. Numerical experiments showcase the performance of each case, respectively. Finally, we investigate the application of TL between two different optimization problems of joint resource allocation (channel assignment and power allocation) in underlay D2D communication. The resource allocation model trained on the dataset obtained from the perfect channel state information (CSI) scenario is transferred to the imperfect CSI scenario and then fine-tuned. The experiment shows that TL improves the performance of the imperfect CSI scenario with less amount of training data.

Se alle arbeider i NVA