Attentive Statistics Pooling for Deep Speaker Embedding
Open accessThe DOI matches its Crossref record; 21 of the 21 references it cites were found in a registry or an index.
Abstract
This paper proposes attentive statistics pooling for deep speaker embedding in text-independent speaker verification. In conventional speaker embedding, frame-level features are averaged over all the frames of a single utterance to form an utterance-level feature. Our method utilizes an attention mechanism to give different weights to different frames and generates not only weighted means but also weighted standard deviations. In this way, it can capture long-term variations in speaker characteristics more effectively. An evaluation on the NIST SRE 2012 and the VoxCeleb data sets shows that it reduces equal error rates (EERs) from the conventional method by 7.5% and 8.1%, respectively.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source says what the claim attributes to it.
In the citing paper
“Attentive Statistics Pooling (ASP) [10] was originally used for extracting speaker embeddings and achieved excellent performance in speaker verification tasks [11].”
From Audio Deepfake Detection With Self-Supervised WavLM And Multi-Fusion Attentive Classifier (2024) · cites this paper as [10]
The cited source says what the claim attributes to it.
In the citing paper
“ASP is a pooling method that combines the advantages of both attention and statistic pooling.”
From Audio Deepfake Detection With Self-Supervised WavLM And Multi-Fusion Attentive Classifier (2024) · cites this paper as [10]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Kaldi Speech Recognition Toolkit
Povey, D. (2024) · Infoscience (Ecole Polytechnique Fédérale de Lausanne) · doi:10.57702/jb3fvbn9 · Cited by 4,609
Front-End Factor Analysis for Speaker Verification
Dehak, N., Kenny, P. J., Dehak, R., Dumouchel, P., & Ouellet, P. (2011) · IEEE Transactions on Audio, Speech, and Language Processing · doi:10.1109/tasl.2010.2064307 · Cited by 3,549 · cited 1×
“Speaker recognition has advanced considerably in the last decade with the i-vector paradigm [1], in which a speech utterance or a speaker is represented in the form of a fixed- lowdimensional feature vector.”
BackgroundX-Vectors: Robust DNN Embeddings for Speaker Recognition
Snyder, D., Garcia-Romero, D., Sell, G., Povey, D., & Khudanpur, S. (2018) · 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) · doi:10.1109/icassp.2018.8461375 · Cited by 2,620 · cited 1×
“The network structure, except for its input dimensions, was exactly the same as the one shown in the recipe 1 published in Kaldi’s official repository [18, 19].”
ResultVoxCeleb: A Large-Scale Speaker Identification Dataset
Nagrani, A., Chung, J. S., & Zisserman, A. (2017) · Interspeech 2017 · doi:10.21437/interspeech.2017-950 · Cited by 2,210 · cited 9×
“, a Time-Delay Neural Network (TDNN) [9], Convolutional Neural Network (CNN) [7, 8], LSTM [10, 11], or Gated Recurrent Unit (GRU) [8].”
BackgroundMethodResultInfluentialProbabilistic Linear Discriminant Analysis for Inferences About Identity
Prince, S. J. D., & Elder, J. H. (2007) · 2007 IEEE 11th International Conference on Computer Vision · doi:10.1109/iccv.2007.4409052 · Cited by 977 · cited 3×
“Mean subtraction, whitening, and length normalization [17] were applied to the i-vector as pre-processing steps before sending it to the PLDA, and similarity was then evaluated using a PLDA model with a speaker space of 400 dimensions.”
MethodAnalysis of i-vector length normalization in speaker recognition systems
Garcia-Romero, D., & Espy-Wilson, C. Y. (2011) · Interspeech 2011 · doi:10.21437/interspeech.2011-53 · Cited by 922 · cited 1×
“Mean subtraction, whitening, and length normalization [17] were applied to the i-vector as pre-processing steps before sending it to the PLDA, and similarity was then evaluated using a PLDA model with a speaker space of 400 dimensions.”
MethodDeep Neural Network Embeddings for Text-Independent Speaker Verification
Snyder, D., Garcia-Romero, D., Povey, D., & Khudanpur, S. (2017) · Interspeech 2017 · doi:10.21437/interspeech.2017-620 · Cited by 739 · cited 7×
“Probabilistic linear discriminant analysis (PLDA) [13, 14] can also be used for measuring the distance between two utterance-level features [9, 10].”
BackgroundMethodInfluentialEnd-to-end text-dependent speaker verification
Heigold, G., Moreno, I., Bengio, S., & Shazeer, N. (2016) · 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) · doi:10.1109/icassp.2016.7472652 · Cited by 514 · cited 1×
“In fixed-phrase textdependent speaker verification, an end-to-end neural networkbased method has been proposed [6] in which Long Short-Term Memory (LSTM) with a single output from the last frame is used to obtain utterance-level speaker features, and it has outperformed conventional i-vector extraction.”
MethodA novel scheme for speaker recognition using a phonetically-aware deep neural network
Lei, Y., Scheffer, N., Ferrer, L., & McLaren, M. (2014) · 2014 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) · doi:10.1109/icassp.2014.6853887 · Cited by 496 · cited 1×
“In early studies [2, 3], deep neural networks (DNNs) derived from acoustic models for ASR have been employed as a universal background model (UBM) to provide phoneme posteriors as well as bottleneck features, which are used for, respectively, zeroth- and first-order statistics in i-vector extraction.”
MethodProbabilistic Linear Discriminant Analysis
Ioffe, S. (2006) · Lecture Notes in Computer Science · doi:10.1007/11744085_41 · Cited by 473 · cited 3×
“Mean subtraction, whitening, and length normalization [17] were applied to the i-vector as pre-processing steps before sending it to the PLDA, and similarity was then evaluated using a PLDA model with a speaker space of 400 dimensions.”
Method
Research works printed but not linked: 3
- [7][7] A. Nagrani, J. S. Chung, and A. Zisserman, 'VoxCeleb: A largescale speaker identification dataset,' in Proc. Interspeech , 2017, pp. 2616-2620. -
- [15][15] G. Bhattacharya, J. Alam, T. Stafylakis, and P. Kenny, 'Deep neural network based text-dependent speaker recognition: Preliminary results,' in Proc. Odyssey , 2016, pp. 9-15. -
- [20][20] L. Ferrer, H. Bratt, L. Burget, H. Cernocky, O. Glembek, M. Graciarena, A. Lawson, Y. Lei, P. Matejka, O. Plchot et al. , 'Promoting robustness for speaker modeling in the community: the PRISM evaluation set,' in Proceedings of NIST 2011 workshop , 2011. -
Cited by
Desplanques, B., Thienpondt, J., & Demuynck, K. (2020) · Interspeech 2020 · doi:10.21437/interspeech.2020-2650 · Cited by 1,568
“The attention system [7] allows the model to select the frames it deems relevant.”
BackgroundMethodInfluentialEnd-to-End anti-spoofing with RawNet2
Tak, H., Patino, J., Todisco, M., Nautsch, A., Evans, N., & Larcher, A. (2021) · ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) · doi:10.1109/icassp39728.2021.9414234 · Cited by 402
“Experiments using the VoxCeleb1 database show that RawNet1 gives an 11% relative improvement in terms of an equal error rate (EER) metric compared to an i-vector based system [16], and is also shown to be competitive with those of an x-vector system [17].”
MethodIn Defence of Metric Learning for Speaker Recognition
Chung, J. S., Huh, J., Mun, S., Lee, M., Heo, H.-S., Choe, S., Ham, C., Jung, S., Lee, B.-J., & Han, I. (2020) · Interspeech 2020 · doi:10.21437/interspeech.2020-1064 · Cited by 383
“A range of encoders have been proposed to aggregate frame-level informations into utterance-level embeddings, from simple averaging [1] to statistical pooling [4, 7] and dictionary-based encodings [17, 34].”
BackgroundMethodSpeaker recognition based on deep learning: An overview
Bai, Z., & Zhang, X.-L. (2021) · Neural Networks · doi:10.1016/j.neunet.2021.03.004 · Cited by 386
“ss of the orthogonality regularization by TDNN. Generally, TDNN has been frequently used as a framework to study other key components of the deep embedding models, such as the temporal pooling layers [84, 85] and objective functions [86, 87, 88]. The TDNN structure has also been intensively improved. For instance, an…”
BackgroundMethodInfluentialUtterance-level Aggregation for Speaker Recognition in the Wild
Xie, W., Nagrani, A., Chung, J. S., & Zisserman, A. (2019) · ICASSP 2019 - 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) · doi:10.1109/icassp.2019.8683120 · Cited by 327
“[15, 13, 19] do not report results on the VoxCeleb-E and VoxCeleb-H) [3] test sets.”
BackgroundMethodInfluentialAttention, please! A survey of neural attention models in deep learning
de Santana Correia, A., & Colombini, E. L. (2022) · Artificial Intelligence Review · doi:10.1007/s10462-022-10148-x · Cited by 298
Biometrics recognition using deep learning: a survey
Minaee, S., Abdolrashidi, A., Su, H., Bennamoun, M., & Zhang, D. (2023) · Artificial Intelligence Review · doi:10.1007/s10462-022-10237-x · Cited by 245
Jung, J.-W., Heo, H.-S., Kim, J.-H., Shim, H.-J., & Yu, H.-J. (2019) · Interspeech 2019 · doi:10.21437/interspeech.2019-1982 · Cited by 166
“[25] MFCC x-vector PLDA Softmax 1500 Heavy 3.”
BackgroundMethodMSER: Multimodal speech emotion recognition using cross-attention with deep fusion
Khan, M., Gueaieb, W., El Saddik, A., & Kwon, S. (2024) · Expert Systems with Applications · doi:10.1016/j.eswa.2023.122946 · Cited by 161
Exploring wav2vec 2.0 on Speaker Verification and Language Identification
Fan, Z., Li, M., Zhou, S., & Xu, B. (2021) · Interspeech 2021 · doi:10.21437/interspeech.2021-1280 · Cited by 128
“41 attentive statistics [28] 3.”
Margin Matters: Towards More Discriminative Deep Neural Network Embeddings for Speaker Recognition
Xiang, X., Wang, S., Huang, H., Qian, Y., & Yu, K. (2019) · Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) · doi:10.1109/apsipaasc47483.2019.9023039 · Cited by 153
CAM++: A Fast and Efficient Network for Speaker Verification Using Context-Aware Masking
Wang, H., Zheng, S., Chen, Y., Cheng, L., & Chen, Q. (2023) · INTERSPEECH 2023 · doi:10.21437/interspeech.2023-1513 · Cited by 150
MFA-Conformer: Multi-scale Feature Aggregation Conformer for Automatic Speaker Verification
Zhang, Y., Lv, Z., Wu, H., Zhang, S., Hu, P., Wu, Z., Lee, H.-Y., & Meng, H. (2022) · Interspeech 2022 · doi:10.21437/interspeech.2022-563 · Cited by 132
CN-Celeb: Multi-genre speaker recognition
Li, L., Liu, R., Kang, J., Fan, Y. B., Cui, H., Cai, Y., Vipperla, R., Zheng, T. F., & Wang, D. (2022) · Speech Communication · doi:10.1016/j.specom.2022.01.002 · Cited by 162
“Fortunately, the DNN-based methods have shown great potential in dealing with this problem [7, 8, 9, 10].”
BackgroundMethodLarge-Scale Self-Supervised Speech Representation Learning for Automatic Speaker Verification
Chen, Z., Chen, S., Wu, Y., Qian, Y., Wang, C., Liu, S., Qian, Y., & Zeng, M. (2022) · ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) · doi:10.1109/icassp43922.2022.9747814 · Cited by 112
“Researchers have developed various neural network architectures [1, 3, 4, 5], training objectives [6, 7, 8, 9], pooling functions [10, 11] to push the limits of the system performance.”
BackgroundJung, J.-W., Kim, S.-B., Shim, H.-J., Kim, J.-H., & Yu, H.-J. (2020) · Interspeech 2020 · doi:10.21437/interspeech.2020-1011 · Cited by 109
“However, applying an attention mechanism to speaker verification is more focused on attentive pooling than applying such mechanisms to intermediate feature maps in image domain tasks [7, 16, 17].”
MethodClova Baseline System for the VoxCeleb Speaker Recognition Challenge 2020
Heo, H. S., Lee, B.-J., Huh, J., & Chung, J. S. (2020) · arXiv · doi:10.48550/arxiv.2009.14153 · Cited by 90
“We refer to this configuration as H / ASP in the results.”
MethodInfluentialThe Vicomtech Audio Deepfake Detection System Based on Wav2vec2 for the 2022 ADD Challenge
Martin-Donas, J. M., & Alvarez, A. (2022) · ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) · doi:10.1109/icassp43922.2022.9747768 · Cited by 100
Deep Speaker Embedding Learning with Multi-level Pooling for Text-independent Speaker Verification
Tang, Y., Ding, G., Huang, J. Y., He, X., & Zhou, B. (2019) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp.2019.8682712 · Cited by 104
“Okabe et al. [8] proposed parametric based attentive pooling to give different importance for each frame feature.”
Audio Deepfake Detection With Self-Supervised WavLM And Multi-Fusion Attentive Classifier
Guo, Y., Huang, H., Chen, X., Zhao, H., & Wang, Y. (2024) · ICASSP 2024 - 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) · doi:10.1109/icassp48485.2024.10447923 · Cited by 95
“Additionally, we pro-pose the Multi-Fusion Attentive (MFA) classifier based on the attentive statistics pooling layer [10].”
MethodVoiceID Loss: Speech Enhancement for Speaker Verification
Shon, S., Tang, H., & Glass, J. R. (2019) · Interspeech (USB) · doi:10.21437/interspeech.2019-1496 · Cited by 88
“various utterance aggregation approaches [22, 18, 23], which show better performance than the Softmax and Cosine similarity back-end.”
MethodSelf-Attention Encoding and Pooling for Speaker Recognition
Safari, P., India, M., & Hernando, J. (2020) · Interspeech (USB) · doi:10.21437/interspeech.2020-1446 · Cited by 96
“This kind of attention pooling is different from the ones proposed in [14, 36], where attention is applied to extract statistics necessary for statistical pooling.”
Meta-Learning for Short Utterance Speaker Recognition with Imbalance Length Pairs
Kye, S. M., Jung, Y., Lee, H., Hwang, S. J., & Kim, H. (2020) · Interspeech (USB) · doi:10.21437/interspeech.2020-1283 · Cited by 84
“For fair comparison, we report baselines without VAD and data augmentation except for x-vector [14] based models [18].”
MethodInfluentialBiometric Recognition Using Deep Learning: A SurveyPreprint
Minaee, S., Abdolrashidi, A., Su, H., Bennamoun, M., & Zhang, D. D. (2019) · arXiv (Cornell University) · Cited by 79
A Deep Neural Network for Short-Segment Speaker Recognition
Hajavi, A., & Etemad, A. (2019) · Interspeech (USB) · doi:10.21437/interspeech.2019-2240 · Cited by 77
“As illustrated in this table, UtterIdNet is closely outperformed by [29], [3], and [22].”
BackgroundMethodInfluentialShi, X. J., Yu, F., Lu, Y., Liang, Y., Feng, Q., Wang, D., Qian, Y., & Xie, L. (2021) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp39728.2021.9413386 · Cited by 108
“Accent recognition is similar to language identification [5, 6, 7] and speaker identification [8, 9, 10, 11].”
BackgroundDeepfake audio detection by speaker verification
Pianese, A., Cozzolino, D., Poggi, G., & Verdoliva, L. (2022) · 2022 IEEE International Workshop on Information Forensics and Security (WIFS) · doi:10.1109/wifs55849.2022.9975428 · Cited by 74
“Attentive Statistic Pooling (ASP) [22] is used, where the channel-wise weighted standard deviation concurs with the weighted mean for temporal-frame aggregation.”
MethodInfluentialThe CORAL+ Algorithm for Unsupervised Domain Adaptation of PLDA
Lee, K. A., Wang, Q., & Koshinaka, T. (2019) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp.2019.8682852 · Cited by 92
“A slight difference here is that we used an attention model in the pooling layer and extended the data augmentation [21].”
MethodSelf-citationGraph Attention Networks for Anti-Spoofing
Tak, H., Jung, J.-W., Patino, J., Todisco, M., & Evans, N. (2021) · Interspeech 2021 · doi:10.21437/interspeech.2021-993 · Cited by 79
“We used the ResNet-18 systems in order to compare the bene-fit of GAT-based attention to alternative attention mechanisms: statistics pooling (SP) [37]; self-attentive pooling (SAP) which assigns different weights to different frames using a weighted mean; attentive statistical pooling (ASP) [38] which generates differ…”
MethodXia, W., Zhang, C., Weng, C., Yu, M., & Yu, D. (2021) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp39728.2021.9414973 · Cited by 92
“New temporal pooling methods [12, 13] were presented to aggregate a variable-length audio input to a fixed length utterance level representation.”
MethodFrequency and Temporal Convolutional Attention for Text-Independent Speaker Recognition
Yadav, S., & Rai, A. (2020) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp40776.2020.9054440 · Cited by 62
“[11] combined attention mechanism with statistics pooling [5] to propose attentive-statistics pooling.”
BackgroundFilteraugment: An Acoustic Environmental Data Augmentation Method
Nam, H., Kim, S.-H., & Park, Y. (2022) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp43922.2022.9747680 · Cited by 97
Simple Attention Module Based Speaker Verification with Iterative Noisy Label Detection
Qin, X., Li, N., Weng, C., Su, D., & Li, M. (2022) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp43922.2022.9746294 · Cited by 91
“The encoding layer is based on global statistic pooling (GSP) or attentive statistics pooling (ASP) [24].”
BackgroundCNN with Phonetic Attention for Text-Independent Speaker Verification
Zhou, T., Zhao, Y. H., Li, J., Gong, Y., & Wu, J. (2019) · IEEE Workshop on Automatic Speech Recognition and Understanding (ASRU) · doi:10.1109/asru46091.2019.9003826 · Cited by 64
Han, Z., Zhuo, Y., Yan, Y., Jin, H., & Fu, H. (2022) · Mechanical Systems and Signal Processing · doi:10.1016/j.ymssp.2022.109367 · Cited by 64
Utterance-level End-to-end Language Identification Using Attention-based CNN-BLSTM
Cai, W., Cai, D., Huang, S., & Li, M. (2019) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp.2019.8682386 · Cited by 68
“To handle the variable-length speech utterances, recently, several context-independent pooling layers such as temporal average pooling (TAP) [10, 11] layer, self-attentive pooling (SAP) layer [12, 13], and learnable dictionary encoding (LDE) [14] layer has been introduced for the utterance-level LID modeling.”
BackgroundThienpondt, J., Desplanques, B., & Demuynck, K. (2021) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp39728.2021.9414600 · Cited by 114
“The statistics pooling layer that maps the variable length input to a fixedlength representation can be enhanced through a temporal attention mechanism [6, 7].”
BackgroundSpeaker Augmentation and Bandwidth Extension for Deep Speaker Embedding
Yamamoto, H., Lee, K. A., Okabe, K., & Koshinaka, T. (2019) · Interspeech (USB) · doi:10.21437/interspeech.2019-1508 · Cited by 81
Self-citationA Gated Recurrent Convolutional Neural Network for Robust Spoofing Detection
Gomez-Alanis, A., Peinado, A. M., González, J. A., & Gomez, A. M. (2019) · IEEE/ACM Transactions on Audio Speech and Language Processing · doi:10.1109/taslp.2019.2937413 · Cited by 60
“1, such as averaging [18], attentive statistics pooling [19], or by using recurrent neural networks (RNNs) [20].”
MethodHow to Improve Your Speaker Embeddings Extractor in Generic Toolkits
Zeinali, H., Burget, L., Rohdin, J., Stafylakis, T., & Černocký, J. (2019) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp.2019.8683445 · Cited by 47
“values) using suggested formulas in [28].”
ASV-SUBTOOLS: Open Source Toolkit for Automatic Speaker Verification
Tong, F., Zhao, M., Zhou, J., Lu, H., Li, Z., Li, L., & Hong, Q. (2021) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp39728.2021.9414676 · Cited by 60
“…architectures (e.g., TDNN [2], F-TDNN [3], ResNet), pooling layers (e.g., learnable dictionary encoder pooling [4], attentive statistic pooling [5], multi-head attention pooling [6]), cost functions (e.g., AM-softmax [7], AAM-softmax [8]), regularization strategies (e.g., L2 regularization,…”
BackgroundLi, D., Liu, F., Shen, T., Chen, L., & Zhao, D. (2023) · Applied Acoustics · doi:10.1016/j.apacoust.2023.109344 · Cited by 59
A Large-Scale Evaluation of Speech Foundation Models
Yang, S.-W., Chang, H.-J., Huang, Z., Liu, A. T., Lai, C.-I., Wu, H., Shi, J., Chang, X., Tsai, H.-S., Huang, W.-C., et al. (2024) · IEEE/ACM Transactions on Audio Speech and Language Processing · doi:10.1109/taslp.2024.3389631 · Cited by 100
Training Multi-task Adversarial Network for Extracting Noise-robust Speaker Embedding
Zhou, J., Jiang, T., Li, L., Hong, Q., Wang, Z., & Xia, B.-Y. (2019) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp.2019.8683828 · Cited by 45
“More recently, many studies [7, 8, 9] have concentrated on extracting utterance-level representation, which is known as speaker embedding, using neural networks combined with a pooling layer.”
BackgroundThe ins and outs of speaker recognition: lessons from VoxSRC 2020
Kwon, Y., Heo, H.-S., Lee, B., & Chung, J. S. (2021) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp39728.2021.9413948 · Cited by 75
“Table 1 shows the detailed architecture of this model, referred to as H / ASP in the results.”
BackgroundMethodInfluentialSelf-Supervised Learning Based Domain Adaptation for Robust Speaker Verification
Chen, Z., Wang, S., & Qian, Y. (2021) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp39728.2021.9414261 · Cited by 44
“To improve the performance and robustness of speaker verification systems, researchers have designed different network backbones [2, 3, 4], different pooling functions [5, 6, 7] and loss functions [8, 9, 10, 11, 12].”
BackgroundDetecting Fake Audio of Arabic Speakers Using Self-Supervised Deep Learning
Almutairi, Z. M., & ElGibreen, H. (2023) · IEEE Access · doi:10.1109/access.2023.3286864 · Cited by 56
“The resulting context vector from the final projection layer of the pre-trained model is passed after that to the ASP layer with Mean ( µ ) measure [45].”
BackgroundMethodDisentangling Style Factors from Speaker Representations
Williams, J. A., & King, S. (2019) · Interspeech (USB) · doi:10.21437/interspeech.2019-1769 · Cited by 56
Multi-View Self-Attention Based Transformer for Speaker Recognition
Wang, R., Ao, J., Zhou, L., Liu, S., Wei, Z., Ko, T., Li, Q., & Zhang, Y. (2022) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp43922.2022.9746639 · Cited by 57
“[12] propose attentive statistics pooling, which gives different weights to different frames and generates weighted means and standard deviations.”
BackgroundMethodInfluentialJung, Y., Kye, S. M., Choi, Y., Jung, M., & Kim, H. (2020) · Interspeech (USB) · doi:10.21437/interspeech.2020-1025 · Cited by 41
Automatic Modulation Classification with Deep Neural Networks
Harper, C. A., Thornton, M. A., & Larson, E. C. (2023) · Electronics · doi:10.3390/electronics12183962 · Cited by 44
Multi-Resolution Multi-Head Attention in Deep Speaker Embedding
Wang, Z., Yao, K., Li, X., & Fang, S. (2020) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp40776.2020.9053217 · Cited by 41
“(2) is used to compute statistics of weighted mean and standard deviation rather than using α t = 1 /N , which is the widely used attentive statistics pooling [5].”
BackgroundMethodInfluentialDeep speaker embeddings for Speaker Verification: Review and experimental comparison
Jakubec, M., Jarina, R., Lieskovská, E., & Kasák, P. (2023) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2023.107232 · Cited by 62
Deep Normalization for Speaker Vectors
Cai, Y., Li, L., Abel, A. K., Zhu, X., & Wang, D. (2020) · IEEE/ACM Transactions on Audio Speech and Language Processing · doi:10.1109/taslp.2020.3039573 · Cited by 42
“icantly improved by developing a more comprehensive architecture [20], [21], improved pooling methods [17], [22]–[24], training criteria [25]–[30], and training schemes [31]–[33].”
BackgroundBhattacharya, G., Alam, J., & Kenny, P. J. (2019) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp.2019.8682611 · Cited by 41
“Self-Attention models are an active area of research in the speaker verification community [14, 16, 17].”
BackgroundMethodTarget Speaker Verification With Selective Auditory Attention for Single and Multi-Talker Speech
Xu, C., Rao, W., Wu, J., & Li, H. (2021) · IEEE/ACM Transactions on Audio Speech and Language Processing · doi:10.1109/taslp.2021.3100682 · Cited by 34
“In addition, when deriving an utterance-level speaker embedding from a frame-level representation, studies have shown the superiority of using the attention mechanism to replace the mean pooling [35], [36].”
S-Vectors and TESA: Speaker Embeddings and a Speaker Authenticator Based on Transformer Encoder
Mary, N. J. M. S., Umesh, S., & Katta, S. V. (2021) · IEEE/ACM Transactions on Audio Speech and Language Processing · doi:10.1109/taslp.2021.3134566 · Cited by 48
“Its strength is that it is not restricted to finite context and attends to all frames in each of its layers as opposed to [16], [15], [17], [18], where attention is employed only during pooling.”
BackgroundMethodOn Robustness of Unsupervised Domain Adaptation for Speaker Recognition
Bousquet, P.-M., & Rouvier, M. (2019) · Interspeech (USB) · doi:10.21437/interspeech.2019-1524 · Cited by 32
“The xvector extractor is a variant of Kaldi toolkit [2] in which we implemented attentive statistics pooling layer [20] .”
MethodDeep multi-metric learning for text-independent speaker verification
Xu, J., Wang, X., Feng, B., & Liu, W. (2020) · Neurocomputing · doi:10.1016/j.neucom.2020.06.045 · Cited by 36
“70 LM [51] VoxCeleb1+PRISM TDNN Multi-Crop 4.”
MethodInfluentialSelf-Supervised Speaker Verification with Simple Siamese Network and Self-Supervised Regularization
Sang, M., Li, H., Liu, F., Arnold, A. O., & Wan, L. (2022) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp43922.2022.9747526 · Cited by 47
“Although great progress has been made on supervised SV systems with various proposed deep neural network architectures [1, 2, 3], novel loss functions [4, 5, 6, 7], different pooling methods [8, 9], and frameworks for domain mismatch [10, 11, 12], they inevitably rely on the availability of largescale annotated datase…”
BackgroundThe SpeakIn System for VoxCeleb Speaker Recognition Challange 2021Preprint
Zhao, M., Ma, Y., Liu, M., & Xu, M. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2109.01989 · Cited by 74
“An attention mechanism [16] is introduced to calculate weighted statistics of the outputs of the backbone.”
BackgroundAttention Mechanism in Speaker Recognition: What Does it Learn in Deep Speaker Embedding?
Wang, Q., Okabe, K., Lee, K. A., Yamamoto, H., & Koshinaka, T. (2018) · IEEE Spoken Language Technology Workshop (SLT) · doi:10.1109/slt.2018.8639586 · Cited by 30
“In [16], it was shown that applying the attention model to an x-vector extraction network improves speaker verification performance, which indicates that the attention weights are able to represent the importance of deep frame-level features.”
BackgroundMethodInfluentialSelf-citationXi-Vector Embedding for Speaker Recognition
Lee, K. A., Wang, Q., & Koshinaka, T. (2021) · IEEE Signal Processing Letters · doi:10.1109/lsp.2021.3091932 · Cited by 55
“In a way, it plays a role similar to the attention model reported in [14], [15].”
MethodResultSelf-citationCombining Automatic Speaker Verification and Prosody Analysis for Synthetic Speech Detection
Attorresi, L., Salvi, D., Borrelli, C., Bestagini, P., & Tubaro, S. (2023) · Lecture Notes in Computer Science · doi:10.1007/978-3-031-37742-6_21 · Cited by 30
Partially Fake Audio Detection by Self-Attention-Based Fake Span Discovery
Wu, H., Kuo, H.-C., Zheng, N., Hung, K.-H., Lee, H.-Y., Tsao, Y., Wang, H., & Meng, H. M. L. (2022) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp43922.2022.9746162 · Cited by 59
“We fix the FFT window size as 384, apply only ASP pooling, adopt data augmentation and the re-synthesised data.”
InfluentialTriplet Based Embedding Distance and Similarity Learning for Text-independent Speaker Verification
Ren, Z., Chen, Z., & Xu, S. (2019) · Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) · doi:10.1109/apsipaasc47483.2019.9023253 · Cited by 31
“Other pooling methods such as self-attentive pooling [4], attentive statistical pooling [5] and high-order statistical pooling [6] also show their advantages.”
MethodA multimodal teacher speech emotion recognition method in the smart classroom
Zhao, G. L., Zhang, Y., & Chu, J. (2024) · Internet of Things · doi:10.1016/j.iot.2024.101069 · Cited by 38
Spatial Pyramid Encoding with Convex Length Normalization for Text-Independent Speaker Verification
Jung, Y., Kim, Y., Lim, H., Choi, Y., & Kim, H. (2019) · Interspeech (USB) · doi:10.21437/interspeech.2019-2177 · Cited by 28
“[16] combined the attention mechanism and the statistics pooling layer to propose attentive statistics pooling layer.”
BackgroundMethodFoolHD: Fooling Speaker Identification by Highly Imperceptible Adversarial Disturbances
Shamsabadi, A. S., Teixeira, F. S., Abad, A., Raj, B., Cavallaro, A., & Trancoso, I. M. (2021) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp39728.2021.9413760 · Cited by 34
“In the second block, an attentive statistical pooling layer [27] weighs each time-frame according to its importance and then computes utterance-level statistics (mean and standard deviation) across the time dimension, providing a summary for the entire speech file.”
BackgroundKoluguri, N. R., Li, J., Lavrukhin, V., & Ginsburg, B. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2010.12653 · Cited by 53
“One can add soft voice activity detection (VAD) or the self-attention mechanism to calculate the weights for frame-level feature vectors, see [15], [16], [17], [18], [19].”
BackgroundCai, D., & Li, M. (2024) · IEEE/ACM Transactions on Audio Speech and Language Processing · doi:10.1109/taslp.2024.3419426 · Cited by 42
Graph Attention Networks for Speaker Verification
Jung, J.-W., Heo, H.-S., Yu, H.-J., & Chung, J. S. (2021) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp39728.2021.9414057 · Cited by 43
“In the ASP layer, the channel-wise weighted standard deviation is calculated in addition to the weighted mean.”
MethodEfficient Self-Supervised Learning Representations for Spoken Language Identification
Liu, H., Garcia, L. P., Khong, A. W. H., Chng, E. S., Styles, S. J., & Khudanpur, S. P. (2022) · IEEE Journal of Selected Topics in Signal Processing · doi:10.1109/jstsp.2022.3201445 · Cited by 40
“To overcome this problem, we propose an attentive SE to perform the dimension-wise scaling on the wav2vec 2.0 speech representations with reference to the attentive statistics pooling [30].”
MethodInfluentialNeXt-TDNN: Modernizing Multi-Scale Temporal Convolution Backbone for Speaker Verification
Heo, H.-J., Shin, U.-H., Lee, R., Cheon, Y., & Park, H. (2024) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp48485.2024.10447037 · Cited by 43
“To extract an utterance-level embedding from the frame-level features, we utilized the ASP pooling layer [20] with channel-dependent attention values α t ∈ R C MFA × 1 following the ECAPA-TDNN.”
MethodGolden Gemini is All You Need: Finding the Sweet Spots for Speaker Verification
Liu, T., Lee, K. A., Wang, Q., & Li, H. (2024) · IEEE/ACM Transactions on Audio Speech and Language Processing · doi:10.1109/taslp.2024.3385277 · Cited by 43
“Commonly used temporal aggregation techniquesincludeaveragepooling[11],statisticalpooling [12], attentive pooling [13], [14], and posterior inference [15].”
Attention Back-End for Automatic Speaker Verification with Multiple Enrollment Utterances
Zeng, C., Wang, X., Cooper, E., Miao, X., & Yamagishi, J. (2022) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp43922.2022.9746688 · Cited by 36
“We also combined our attention back-end with TDNN-ASP and ECAPA-TDNN models, which contained frame-level attention module.”
MethodResultAudio Deepfake Detection System with Neural Stitching for ADD 2022
Yan, R., Cheng, W. M., Zhou, S., Guo, T., Zou, W., & Li, X. (2022) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp43922.2022.9746820 · Cited by 38
C3-DINO: Joint Contrastive and Non-Contrastive Self-Supervised Learning for Speaker Verification
Zhang, C., & Yu, D. (2022) · IEEE Journal of Selected Topics in Signal Processing · doi:10.1109/jstsp.2022.3198315 · Cited by 33
“To aggregate the frame-level information to the utterance-level, different temporal pooling methods [12], [19], [20], [21] were explored as an important step to normalize phonetic variability.”
MethodThe VoxCeleb Speaker Recognition Challenge: A Retrospective
Huh, J., Chung, J. S., Nagrani, A., Brown, A., Jung, J.-W., Garcia-Romero, D., & Zisserman, A. (2024) · IEEE/ACM Transactions on Audio Speech and Language Processing · doi:10.1109/taslp.2024.3444456 · Cited by 43
“All baseline models are ResNetSE34V2 [35] with ASP pooling [100] and are trained with a combination of angular prototypical loss [32] and cross-entropy loss.”
MethodCam: Context-Aware Masking for Robust Speaker Verification
Yu, Y.-Q., Zheng, S., Suo, H., Lei, Y., & Li, W.-J. (2021) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp39728.2021.9414704 · Cited by 36
“Attentive statistics pooling (ASP) [23] is an attention-based pooling strategy.”
BackgroundKim, J.-H., Shim, H.-J., Heo, J., & Yu, H.-J. (2022) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp43922.2022.9747594 · Cited by 24
“Finally, speaker embeddings with 512 dimensions are extracted through the attentive statistical pooling (ASP) [19] and fully connected (FC) layers.”
MethodExtended U-Net for Speaker Verification in Noisy Environments
Kim, J.-H., Heo, J., Shim, H.-J., & Yu, H.-J. (2022) · Interspeech (USB) · doi:10.21437/interspeech.2022-155 · Cited by 28
Multi-Frequency Information Enhanced Channel Attention Module for Speaker Representation Learning
Sang, M., & Hansen, J. H. L. (2022) · Interspeech (USB) · doi:10.21437/interspeech.2022-892 · Cited by 27
Meta-Generalization for Domain-Invariant Speaker Verification
Zhang, H., Wang, L., Lee, K. A., Liu, M., Dang, J., & Meng, H. M. L. (2023) · IEEE/ACM Transactions on Audio Speech and Language Processing · doi:10.1109/taslp.2023.3244518 · Cited by 30
“In [25], [26], [27], the representation ability of embeddings is improved by focusing on the important frames of utterances.”
BackgroundMethodZheng, Q., Chen, Z., Liu, H., Lu, Y., Li, J., & Liu, T. (2022) · Expert Systems with Applications · doi:10.1016/j.eswa.2023.119511 · Cited by 24
Speaker Verification by Partial AUC Optimization With Mahalanobis Distance Metric Learning
Bai, Z., Zhang, X.-L., & Chen, J. (2020) · IEEE/ACM Transactions on Audio Speech and Language Processing · doi:10.1109/taslp.2020.2990275 · Cited by 23
“Attention mechanism was also studied to generate utterance-level features [16], [17].”
BackgroundThe DKU-DukeECE Systems for VoxCeleb Speaker Recognition Challenge 2020Preprint
Wang, W., Cai, D., Qin, X., & Li, M. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2010.12731 · Cited by 49
“The front-end feature extractor is followed by an attentive statistics pooling (ASP) layer [13] that calculates the mean and standard deviations of the final frame-level features.”
MethodCross-Modal Distillation for Speaker Recognition
Jin, Y., Hu, G., Chen, H., Miao, D., Hu, L., & Zhao, C. (2023) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v37i11.26525 · Cited by 30
“All rights reserved. deep learning, neural networks (Nagrani, Chung, and Zisserman 2017; Snyder et al. 2018; Okabe, Koshinaka, and Shinoda 2018) started to become the mainstream and achieved promising progress.”
BackgroundMethodA Federated Attention-Based Multimodal Biometric Recognition Approach in IoT
Lin, L., Zhao, Y., Meng, J., & Zhao, Q. (2023) · Sensors · doi:10.3390/s23136006 · Cited by 24
“For image input, our feature extraction network is based on the ResNet-34 [26] architecture, we integrated Convolutional Block Attention Module (CBAM) [27] into each residual block of ResNet-34 to combine channel attention and spatial attention.”
Voting for the Right Answer: Adversarial Defense for Speaker Verification
Wu, H., Zhang, Y., Wu, Z., Wang, D., & Lee, H.-Y. (2021) · Interspeech (USB) · doi:10.21437/interspeech.2021-1452 · Cited by 30
“The performance with SAP exhibits a similar trend as with ASP , and due to limited space, we only report the results of ASP .”
MethodInfluentialDouble Multi-Head Attention for Speaker Verification
India, M., Safari, P., & Hernando, J. (2021) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp39728.2021.9414877 · Cited by 23
“In works like [19], attention is used to extract better order features statistics.”
MethodImproving Transformer-Based Networks with Locality for Automatic Speaker Verification
Sang, M., Zhao, Y., Liu, G., Hansen, J. H. L., & Wu, J. (2023) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp49357.2023.10096333 · Cited by 27
“Attentive statistic pooling (ASP) [16] is used to generate utterance-level embeddings.”
MethodDilated residual networks with multi-level attention for speaker verification
Wu, Y., Guo, C., Gao, H., Xu, J., & Bai, G. (2020) · Neurocomputing · doi:10.1016/j.neucom.2020.06.079 · Cited by 19
Siamese Capsule Network for End-to-End Speaker Recognition in the Wild
Hajavi, A., & Etemad, A. (2021) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp39728.2021.9414722 · Cited by 19
“[29] TDNN (X-Vector) + TAP Softmax VoxCeleb1 3.”
Pretraining Conformer with ASR for Speaker Verification
Cai, D., Wang, W., Li, M., Xia, R., & Huang, C. (2023) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp49357.2023.10096659 · Cited by 25
“Various network architecture [3, 4, 5, 6], training objectives [7, 8, 9], and training strategies [10, 11] are proposed for speaker verification systems.”
BackgroundMa, B., Kong, Q., Ding, Y., Chen, L., & Gao, W. (2023) · NDT & E International · doi:10.1016/j.ndteint.2023.102986 · Cited by 22
You, L., Guo, W., Dai, L.-R., & Du, J. (2019) · Interspeech (USB) · doi:10.21437/interspeech.2019-2264 · Cited by 23
“In the original x-vector system, low-order statistics, such as the mean and standard deviation, are calculated to perform the statistics pooling and have demonstrated their effectiveness for extracting speaker-specific embeddings [5, 17].”
MethodNEC-TT System for Mixed-Bandwidth and Multi-Domain Speaker Recognition
Lee, K. A., Yamamoto, H., Okabe, K., Wang, Q., Guo, L., Koshinaka, T., Zhang, J., & Shinoda, K. (2019) · Computer Speech & Language · doi:10.1016/j.csl.2019.101033 · Cited by 21
Self-citationA Unified Deep Learning Framework for Short-Duration Speaker Verification in Adverse Environments
Jung, Y., Choi, Y., Lim, H., & Kim, H. (2020) · IEEE Access · doi:10.1109/access.2020.3025941 · Cited by 18
“We followed the same TDNN architecture as in [45].”
BackgroundMethodInfluentialCombination of deep speaker embeddings for diarisation
Sun, G., Zhang, C., & Woodland, P. C. (2021) · Neural Networks · doi:10.1016/j.neunet.2021.04.020 · Cited by 19
“…deviation (Garcia-Romero et al., 2017; Diez et al., 2019; Wang et al., 2018c), attention mechanisms (Chowdhury et al., 2018; Zhu et al., 2018; Sun et al., 2019; Shi et al., 2020), and their combination (Okabe et al., 2018), have been used, which also enables joint training over entire segments.”
Background
Metadata
- Authors
- Okabe, K., Koshinaka, T., & Shinoda, K.
- Affiliations
- NEC (Japan) (JP)
- Tokyo Institute of Technology (JP)
- Venue
- Interspeech 2018
- Publisher
- ISCA
- Published
- 2018-09-02
- Pages
- 2252-2256
- ISSN
- 2308-457X (ISSN-L)
- OpenAlex
- W2794506738
- Semantic Scholar
- 4a64008cd21488071ac418aad32ce7129408edb4
- Cited by
- 514
- References
- 20
- Citations by year
- 2026: 44 · 2025: 47 · 2024: 72 · 2023: 74 · 2022: 90 · 2021: 83 · 2020: 65 · 2019: 37 · 2018: 2
- FWCI
- 19.09
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- Interspeech (USB) (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- Institutional Repositories DataBase (IRDB) (submittedVersion)
- Tokyo Tech Research Repository (Tokyo Institute of Technology) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Speech Recognition and Synthesis
- Physical Sciences › Computer Science › Signal Processing › Speech and Audio Processing
- Physical Sciences › Computer Science › Signal Processing › Biometric Identification and Security
- Keywords
- speaker embedding, text-independent speaker verification, attention mechanism, utterance-level features, frame-level features, VoxCeleb
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.