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.”
BackgroundMethodInfluential
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.