Pub Date : 2025-01-30DOI: 10.1109/OJSP.2025.3536850
David Uliel;Raja Giryes
In unsupervised domain adaptation (UDA) labeled data is available for one domain (Source Domain) which is generated according to some distribution, and unlabeled data is available for a second domain (Target Domain) which is generated from a possibly different distribution but has the same task. The goal is to learn a model that performs well on the target domain although labels are available only for the source data. Many recent works attempt to align the source and the target domains by matching their marginal distributions in a learned feature space. In this paper, we address the domain difference as a nuisance, and enables better adaptability of the domains, by encouraging minimality of the target domain representation, disentanglement of the features, and a smoother feature space that cluster better the target data. To this end, we use the information bottleneck theory and a classical technique from the blind source separation framework, namely, ICA (independent components analysis). We show that these concepts can improve performance of leading domain adaptation methods on various domain adaptation benchmarks.
{"title":"Task Nuisance Filtration for Unsupervised Domain Adaptation","authors":"David Uliel;Raja Giryes","doi":"10.1109/OJSP.2025.3536850","DOIUrl":"https://doi.org/10.1109/OJSP.2025.3536850","url":null,"abstract":"In unsupervised domain adaptation (UDA) labeled data is available for one domain (Source Domain) which is generated according to some distribution, and unlabeled data is available for a second domain (Target Domain) which is generated from a possibly different distribution but has the same task. The goal is to learn a model that performs well on the target domain although labels are available only for the source data. Many recent works attempt to align the source and the target domains by matching their marginal distributions in a learned feature space. In this paper, we address the domain difference as a nuisance, and enables better adaptability of the domains, by encouraging minimality of the target domain representation, disentanglement of the features, and a smoother feature space that cluster better the target data. To this end, we use the information bottleneck theory and a classical technique from the blind source separation framework, namely, ICA (independent components analysis). We show that these concepts can improve performance of leading domain adaptation methods on various domain adaptation benchmarks.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"6 ","pages":"303-311"},"PeriodicalIF":2.9,"publicationDate":"2025-01-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10858365","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143489277","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-01-30DOI: 10.1109/OJSP.2025.3536853
Masahiro Kada;Ryota Yoshihashi;Satoshi Ikehata;Rei Kawakami;Ikuro Sato
Mixture of experts with a sparse expert selection rule has been gaining much attention recently because of its scalability without compromising inference time. However, unlike standard neural networks, sparse mixture-of-experts models inherently exhibit discontinuities in the output space, which may impede the acquisition of appropriate invariance to the input perturbations, leading to a deterioration of model performance for tasks such as classification. To address this issue, we propose Pairwise Router Consistency (PRC) that effectively penalizes the discontinuities occurring under natural deformations of input images. With the supervised loss, the use of PRC loss empirically improves classification accuracy on ImageNet-1 K, CIFAR-10, and CIFAR-100 datasets, compared to a baseline method. Notably, our method with 1-expert selection slightly outperforms the baseline method using 2-expert selection. We also confirmed that models trained with our method experience discontinuous changes less frequently under input perturbations.
{"title":"Robustifying Routers Against Input Perturbations for Sparse Mixture-of-Experts Vision Transformers","authors":"Masahiro Kada;Ryota Yoshihashi;Satoshi Ikehata;Rei Kawakami;Ikuro Sato","doi":"10.1109/OJSP.2025.3536853","DOIUrl":"https://doi.org/10.1109/OJSP.2025.3536853","url":null,"abstract":"Mixture of experts with a sparse expert selection rule has been gaining much attention recently because of its scalability without compromising inference time. However, unlike standard neural networks, sparse mixture-of-experts models inherently exhibit discontinuities in the output space, which may impede the acquisition of appropriate invariance to the input perturbations, leading to a deterioration of model performance for tasks such as classification. To address this issue, we propose Pairwise Router Consistency (PRC) that effectively penalizes the discontinuities occurring under natural deformations of input images. With the supervised loss, the use of PRC loss empirically improves classification accuracy on ImageNet-1 K, CIFAR-10, and CIFAR-100 datasets, compared to a baseline method. Notably, our method with 1-expert selection slightly outperforms the baseline method using 2-expert selection. We also confirmed that models trained with our method experience discontinuous changes less frequently under input perturbations.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"6 ","pages":"276-283"},"PeriodicalIF":2.9,"publicationDate":"2025-01-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10858379","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143465817","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-01-28DOI: 10.1109/OJSP.2025.3534686
Junghyun Koo;Gordon Wichern;François G. Germain;Sameer Khurana;Jonathan Le Roux
We introduce Self-Monitored Inference-Time INtervention (SMITIN), an approach for controlling an autoregressive generative music transformer using classifier probes. These simple logistic regression probes are trained on the output of each attention head in the transformer using a small dataset of audio examples both exhibiting and missing a specific musical trait (e.g., the presence/absence of drums, or real/synthetic music). We then steer the attention heads in the probe direction, ensuring the generative model output captures the desired musical trait. Additionally, we monitor the probe output to avoid adding an excessive amount of intervention into the autoregressive generation, which could lead to temporally incoherent music. We validate our results objectively and subjectively for both audio continuation and text-to-music applications, demonstrating the ability to add controls to large generative models for which retraining or even fine-tuning is impractical for most musicians. Audio samples of the proposed intervention approach are available on our demo page.
{"title":"SMITIN: Self-Monitored Inference-Time INtervention for Generative Music Transformers","authors":"Junghyun Koo;Gordon Wichern;François G. Germain;Sameer Khurana;Jonathan Le Roux","doi":"10.1109/OJSP.2025.3534686","DOIUrl":"https://doi.org/10.1109/OJSP.2025.3534686","url":null,"abstract":"We introduce Self-Monitored Inference-Time INtervention (SMITIN), an approach for controlling an autoregressive generative music transformer using classifier probes. These simple logistic regression probes are trained on the output of each attention head in the transformer using a small dataset of audio examples both exhibiting and missing a specific musical trait (e.g., the presence/absence of drums, or real/synthetic music). We then steer the attention heads in the probe direction, ensuring the generative model output captures the desired musical trait. Additionally, we monitor the probe output to avoid adding an excessive amount of intervention into the autoregressive generation, which could lead to temporally incoherent music. We validate our results objectively and subjectively for both audio continuation and text-to-music applications, demonstrating the ability to add controls to large generative models for which retraining or even fine-tuning is impractical for most musicians. Audio samples of the proposed intervention approach are available on our <underline>demo page</u>.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"6 ","pages":"266-275"},"PeriodicalIF":2.9,"publicationDate":"2025-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10856829","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143465815","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-01-27DOI: 10.1109/OJSP.2025.3534690
Yaman Kındap;Simon Godsill
Probabilistic dynamical models used in applications in tracking and prediction are typically assumed to be Gaussian noise driven motions since well-known inference algorithms can be applied to these models. However, in many real world examples deviations from Gaussianity are expected to appear, e.g., rapid changes in speed or direction, which cannot be reflected using processes with a smooth mean response. In this work, we introduce the non-Gaussian process (NGP) dynamical model which allow for straightforward modelling of heavy-tailed, non-Gaussian behaviours while retaining a tractable conditional Gaussian process (GP) structure through an infinite mixture of non-homogeneous GPs representation. We present two novel inference methodologies for these new models based on the conditionally Gaussian formulation of NGPs which are suitable for both MCMC and marginalised particle filtering algorithms. The results are demonstrated on synthetically generated data sets.
{"title":"Non-Gaussian Process Dynamical Models","authors":"Yaman Kındap;Simon Godsill","doi":"10.1109/OJSP.2025.3534690","DOIUrl":"https://doi.org/10.1109/OJSP.2025.3534690","url":null,"abstract":"Probabilistic dynamical models used in applications in tracking and prediction are typically assumed to be Gaussian noise driven motions since well-known inference algorithms can be applied to these models. However, in many real world examples deviations from Gaussianity are expected to appear, e.g., rapid changes in speed or direction, which cannot be reflected using processes with a smooth mean response. In this work, we introduce the non-Gaussian process (NGP) dynamical model which allow for straightforward modelling of heavy-tailed, non-Gaussian behaviours while retaining a tractable conditional Gaussian process (GP) structure through an infinite mixture of non-homogeneous GPs representation. We present two novel inference methodologies for these new models based on the conditionally Gaussian formulation of NGPs which are suitable for both MCMC and marginalised particle filtering algorithms. The results are demonstrated on synthetically generated data sets.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"6 ","pages":"213-221"},"PeriodicalIF":2.9,"publicationDate":"2025-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10854574","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143455222","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-01-27DOI: 10.1109/OJSP.2025.3534692
Samuel Rey;Bishwadeep Das;Elvin Isufi
This paper addresses the problem of online network topology inference for expanding graphs from a stream of spatiotemporal signals. Online algorithms for dynamic graph learning are crucial in delay-sensitive applications or when changes in topology occur rapidly. While existing works focus on inferring the connectivity within a fixed set of nodes, in practice, the graph can grow as new nodes join the network. This poses additional challenges like modeling temporal dynamics involving signals and graphs of different sizes. This growth also increases the computational complexity of the learning process, which may become prohibitive. To the best of our knowledge, this is the first work to tackle this setting. We propose a general online algorithm based on projected proximal gradient descent that accounts for the increasing graph size at each iteration. Recursively updating the sample covariance matrix is a key aspect of our approach. We introduce a strategy that enables different types of updates for nodes that just joined the network and for previously existing nodes. To provide further insights into the proposed method, we specialize it in Gaussian Markov random field settings, where we analyze the computational complexity and characterize the dynamic cumulative regret. Finally, we demonstrate the effectiveness of the proposed approach using both controlled experiments and real-world datasets from epidemic and financial networks.
{"title":"Online Learning of Expanding Graphs","authors":"Samuel Rey;Bishwadeep Das;Elvin Isufi","doi":"10.1109/OJSP.2025.3534692","DOIUrl":"https://doi.org/10.1109/OJSP.2025.3534692","url":null,"abstract":"This paper addresses the problem of online network topology inference for expanding graphs from a stream of spatiotemporal signals. Online algorithms for dynamic graph learning are crucial in delay-sensitive applications or when changes in topology occur rapidly. While existing works focus on inferring the connectivity within a fixed set of nodes, in practice, the graph can grow as new nodes join the network. This poses additional challenges like modeling temporal dynamics involving signals and graphs of different sizes. This growth also increases the computational complexity of the learning process, which may become prohibitive. To the best of our knowledge, this is the first work to tackle this setting. We propose a general online algorithm based on projected proximal gradient descent that accounts for the increasing graph size at each iteration. Recursively updating the sample covariance matrix is a key aspect of our approach. We introduce a strategy that enables different types of updates for nodes that just joined the network and for previously existing nodes. To provide further insights into the proposed method, we specialize it in Gaussian Markov random field settings, where we analyze the computational complexity and characterize the dynamic cumulative regret. Finally, we demonstrate the effectiveness of the proposed approach using both controlled experiments and real-world datasets from epidemic and financial networks.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"6 ","pages":"247-255"},"PeriodicalIF":2.9,"publicationDate":"2025-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10854617","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143471129","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-01-20DOI: 10.1109/OJSP.2025.3532199
Zoltan Rozsa;Akos Madaras;Tamas Sziranyi
LiDAR point clouds are a rich source of information for autonomous vehicles and ADAS systems. However, they can be challenging to segment for moving objects as - among other things - finding correspondences between sparse point clouds of consecutive frames is difficult. Traditional methods rely on a (global or local) map of the environment, which can be demanding to acquire and maintain in real-world conditions and the presence of the moving objects themselves. This paper proposes a novel approach using as minimal sweeps as possible to decrease the computational burden and achieve mapless moving object segmentation (MOS) in LiDAR point clouds. Our approach is based on a multimodal learning model with single-modal inference. The model is trained on a dataset of LiDAR point clouds and related camera images. The model learns to associate features from the two modalities, allowing it to predict dynamic objects even in the absence of a map and the camera modality. We propose semantic information usage for multi-frame instance segmentation in order to enhance performance measures. We evaluate our approach to the SemanticKITTI and Apollo real-world autonomous driving datasets. Our results show that our approach can achieve state-of-the-art performance on moving object segmentation and utilize only a few (even one) LiDAR frames.
{"title":"Efficient Moving Object Segmentation in LiDAR Point Clouds Using Minimal Number of Sweeps","authors":"Zoltan Rozsa;Akos Madaras;Tamas Sziranyi","doi":"10.1109/OJSP.2025.3532199","DOIUrl":"https://doi.org/10.1109/OJSP.2025.3532199","url":null,"abstract":"LiDAR point clouds are a rich source of information for autonomous vehicles and ADAS systems. However, they can be challenging to segment for moving objects as - among other things - finding correspondences between sparse point clouds of consecutive frames is difficult. Traditional methods rely on a (global or local) map of the environment, which can be demanding to acquire and maintain in real-world conditions and the presence of the moving objects themselves. This paper proposes a novel approach using as minimal sweeps as possible to decrease the computational burden and achieve mapless moving object segmentation (MOS) in LiDAR point clouds. Our approach is based on a multimodal learning model with single-modal inference. The model is trained on a dataset of LiDAR point clouds and related camera images. The model learns to associate features from the two modalities, allowing it to predict dynamic objects even in the absence of a map and the camera modality. We propose semantic information usage for multi-frame instance segmentation in order to enhance performance measures. We evaluate our approach to the SemanticKITTI and Apollo real-world autonomous driving datasets. Our results show that our approach can achieve state-of-the-art performance on moving object segmentation and utilize only a few (even one) LiDAR frames.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"6 ","pages":"118-128"},"PeriodicalIF":2.9,"publicationDate":"2025-01-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10848132","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143379492","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The Multi-speaker, Multi-lingual Indic Text to Speech (TTS) with voice cloning (LIMMITS'24) challenge is organized as part of the ICASSP 2024 signal processing grand challenge. LIMMITS'24 aims at the development of voice cloning for the multi-speaker, multi-lingual Text-to-Speech (TTS) model. Towards this, 80 hours of TTS data has been released in each of Bengali, Chhattisgarhi, English (Indian), and Kannada languages. This is in addition to Telugu, Hindi, and Marathi data released during the LIMMITS'23 challenge. The challenge encourages the advancement of TTS in Indian Languages as well as the development of multi-speaker voice cloning techniques for TTS. The three tracks of LIMMITS'24 have provided an opportunity for various researchers and practitioners around the world to explore the state of the art in research for voice cloning with TTS.
{"title":"LIMMITS'24: Multi-Speaker, Multi-Lingual INDIC TTS With Voice Cloning","authors":"Sathvik Udupa;Jesuraja Bandekar;Abhayjeet Singh;Deekshitha G;Saurabh Kumar;Sandhya Badiger;Amala Nagireddi;Roopa R;Prasanta Kumar Ghosh;Hema A. Murthy;Pranaw Kumar;Keiichi Tokuda;Mark Hasegawa-Johnson;Philipp Olbrich","doi":"10.1109/OJSP.2025.3531782","DOIUrl":"https://doi.org/10.1109/OJSP.2025.3531782","url":null,"abstract":"The Multi-speaker, Multi-lingual Indic Text to Speech (TTS) with voice cloning (LIMMITS'24) challenge is organized as part of the ICASSP 2024 signal processing grand challenge. LIMMITS'24 aims at the development of voice cloning for the multi-speaker, multi-lingual Text-to-Speech (TTS) model. Towards this, 80 hours of TTS data has been released in each of Bengali, Chhattisgarhi, English (Indian), and Kannada languages. This is in addition to Telugu, Hindi, and Marathi data released during the LIMMITS'23 challenge. The challenge encourages the advancement of TTS in Indian Languages as well as the development of multi-speaker voice cloning techniques for TTS. The three tracks of LIMMITS'24 have provided an opportunity for various researchers and practitioners around the world to explore the state of the art in research for voice cloning with TTS.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"6 ","pages":"293-302"},"PeriodicalIF":2.9,"publicationDate":"2025-01-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10845816","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143489278","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Sign Language conveys information through multiple channels composed of manual (handshape, hand movement) and non-manual (facial expression, mouthing, body posture) components. Sign language assessment involves giving granular feedback to a learner, in terms of correctness of the manual and non-manual components, aiding the learner's progress. Existing methods rely on handcrafted skeleton-based features for hand movement within a KL-HMM framework to identify errors in manual components. However, modern deep learning models offer powerful spatio-temporal representations for videos to represent hand movement and facial expressions. Despite their success in classification tasks, these representations often struggle to attribute errors to specific sources, such as incorrect handshape, improper movement, or incorrect facial expressions. To address this limitation, we leverage and analyze the spatio-temporal representations from Inflated 3D Convolutional Networks (I3D) and integrate them into the KL-HMM framework to assess sign language videos on both manual and non-manual components. By applying masking and cropping techniques, we isolate and evaluate distinct channels of hand movement, and facial expressions using the I3D model and handshape using the CNN-based model. Our approach outperforms traditional methods based on handcrafted features, as validated through experiments on the SMILE-DSGS dataset, and therefore demonstrates that it can enhance the effectiveness of sign language learning tools.
{"title":"Posterior-Based Analysis of Spatio-Temporal Features for Sign Language Assessment","authors":"Neha Tarigopula;Sandrine Tornay;Ozge Mercanoglu Sincan;Richard Bowden;Mathew Magimai.-Doss","doi":"10.1109/OJSP.2025.3531781","DOIUrl":"https://doi.org/10.1109/OJSP.2025.3531781","url":null,"abstract":"Sign Language conveys information through multiple channels composed of manual (handshape, hand movement) and non-manual (facial expression, mouthing, body posture) components. Sign language assessment involves giving granular feedback to a learner, in terms of correctness of the manual and non-manual components, aiding the learner's progress. Existing methods rely on handcrafted skeleton-based features for hand movement within a KL-HMM framework to identify errors in manual components. However, modern deep learning models offer powerful spatio-temporal representations for videos to represent hand movement and facial expressions. Despite their success in classification tasks, these representations often struggle to attribute errors to specific sources, such as incorrect handshape, improper movement, or incorrect facial expressions. To address this limitation, we leverage and analyze the spatio-temporal representations from Inflated 3D Convolutional Networks (I3D) and integrate them into the KL-HMM framework to assess sign language videos on both manual and non-manual components. By applying masking and cropping techniques, we isolate and evaluate distinct channels of hand movement, and facial expressions using the I3D model and handshape using the CNN-based model. Our approach outperforms traditional methods based on handcrafted features, as validated through experiments on the SMILE-DSGS dataset, and therefore demonstrates that it can enhance the effectiveness of sign language learning tools.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"6 ","pages":"284-292"},"PeriodicalIF":2.9,"publicationDate":"2025-01-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10845152","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143465818","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Correction to “Energy Efficient Signal Detection Using SPRT and Ordered Transmissions in Wireless Sensor Networks”","authors":"Shailee Yagnik;Ramanarayanan Viswanathan;Lei Cao","doi":"10.1109/OJSP.2024.3519916","DOIUrl":"https://doi.org/10.1109/OJSP.2024.3519916","url":null,"abstract":"In [1, p. 1124], a footnote is needed on (13) as shown below: begin{equation*}qquadqquadquad{{alpha }^# } < left( {1 - {{c}_1}} right)alpha + left( {1 - left( {1 - {{c}_1}} right)alpha } right)alphaqquadqquadquad hbox{(13)$^{1}$} end{equation*}","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"6 ","pages":"16-16"},"PeriodicalIF":2.9,"publicationDate":"2025-01-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10845022","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142993599","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Formant tracking is an area of speech science that has recently undergone a technology shift from classical model-driven signal processing methods to modern data-driven deep learning methods. In this study, these two domains are combined in formant tracking by refining the formants estimated by a data-driven deep neural network (DNN) with formant estimates given by a model-driven linear prediction (LP) method. In the refinement process, the three lowest formants, initially estimated by the DNN-based method, are frame-wise replaced with local spectral peaks identified by the LP method. The LP-based refinement stage can be seamlessly integrated into the DNN without any training. As an LP method, the study advocates the use of quasiclosed phase forward-backward (QCP-FB) analysis. Three spectral representations are compared as DNN inputs: mel-frequency cepstral coefficients (MFCCs), the spectrogram, and the complex spectrogram. Formant tracking performance was evaluated by comparing the proposed refined DNN tracker with seven reference trackers, which included both signal processing and deep learning based methods. As evaluation data, ground truth formants of the Vocal Tract Resonance (VTR) corpus were used. The results demonstrate that the refined DNN trackers outperformed all conventional trackers. The best results were obtained by using the MFCC input for the DNN. The proposed MFCC refinement (MFCC-DNNQCP-FB) reduced estimation errors by 0.8 Hz, 12.9 Hz, and 11.7 Hz for the first (F1), second (F2), and third (F3) formants, respectively, compared to the Deep Formants refinement (DeepFQCP-FB). When compared to the model-driven KARMA tracking method, the proposed refinement reduced estimation errors by 2.3 Hz, 55.5 Hz, and 143.4 Hz for F1, F2, and F3, respectively. A detailed evaluation across various phonetic categories and gender groups showed that the proposed hybrid refinement approach improves formanttracking performance across most test conditions.
{"title":"Formant Tracking by Combining Deep Neural Network and Linear Prediction","authors":"Sudarsana Reddy Kadiri;Kevin Huang;Christina Hagedorn;Dani Byrd;Paavo Alku;Shrikanth Narayanan","doi":"10.1109/OJSP.2025.3530876","DOIUrl":"https://doi.org/10.1109/OJSP.2025.3530876","url":null,"abstract":"Formant tracking is an area of speech science that has recently undergone a technology shift from classical model-driven signal processing methods to modern data-driven deep learning methods. In this study, these two domains are combined in formant tracking by refining the formants estimated by a data-driven deep neural network (DNN) with formant estimates given by a model-driven linear prediction (LP) method. In the refinement process, the three lowest formants, initially estimated by the DNN-based method, are frame-wise replaced with local spectral peaks identified by the LP method. The LP-based refinement stage can be seamlessly integrated into the DNN without any training. As an LP method, the study advocates the use of quasiclosed phase forward-backward (QCP-FB) analysis. Three spectral representations are compared as DNN inputs: mel-frequency cepstral coefficients (MFCCs), the spectrogram, and the complex spectrogram. Formant tracking performance was evaluated by comparing the proposed refined DNN tracker with seven reference trackers, which included both signal processing and deep learning based methods. As evaluation data, ground truth formants of the Vocal Tract Resonance (VTR) corpus were used. The results demonstrate that the refined DNN trackers outperformed all conventional trackers. The best results were obtained by using the MFCC input for the DNN. The proposed MFCC refinement (MFCC-DNN<sub>QCP-FB</sub>) reduced estimation errors by 0.8 Hz, 12.9 Hz, and 11.7 Hz for the first (F1), second (F2), and third (F3) formants, respectively, compared to the Deep Formants refinement (DeepF<sub>QCP-FB</sub>). When compared to the model-driven KARMA tracking method, the proposed refinement reduced estimation errors by 2.3 Hz, 55.5 Hz, and 143.4 Hz for F1, F2, and F3, respectively. A detailed evaluation across various phonetic categories and gender groups showed that the proposed hybrid refinement approach improves formanttracking performance across most test conditions.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"6 ","pages":"222-230"},"PeriodicalIF":2.9,"publicationDate":"2025-01-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10843356","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143430569","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：