ACA-Net: Towards Lightweight Speaker Verification using Asymmetric Cross Attention

Jia Qi Yip, Tuan Truong, Dianwen Ng, Chong Zhang, Yukun Ma +5 more
5/20/2023
cs.SDcs.LGeess.AS

No official code found. Best implementation: Yip-Jia-Qi/ACA-Net (8 stars, MIT, last commit 2023-09-07).

Abstract

In this paper, we propose ACA-Net, a lightweight, global context-aware speaker embedding extractor for Speaker Verification (SV) that improves upon existing work by using Asymmetric Cross Attention (ACA) to replace temporal pooling. ACA is able to distill large, variable-length sequences into small, fixed-sized latents by attending a small query to large key and value matrices. In ACA-Net, we build a Multi-Layer Aggregation (MLA) block using ACA to generate fixed-sized identity vectors from variable-length inputs. Through global attention, ACA-Net acts as an efficient global feature extractor that adapts to temporal variability unlike existing SV models that apply a fixed function for pooling over the temporal dimension which may obscure information about the signal's non-stationary temporal variability. Our experiments on the WSJ0-1talker show ACA-Net outperforms a strong baseline by 5\% relative improvement in EER using only 1/5 of the parameters.

View on arXivView PDF

Code Implementations(1)

Yip-Jia-Qi/ACA-NetCommunity100%
MIT

Pytorch Implementation of ACA-Net for Speaker Verification

81PythonMay 20, 20233 years agoMIT

Similar Papers

Speaker Verification with Speech-Aware LLMs: Evaluation and Augmentation

79% match2026

MSA-ASR: Efficient Multilingual Speaker Attribution with frozen ASR Models

79% match2024

Layer-aware TDNN: Speaker Recognition Using Multi-Layer Features from Pre-Trained Models

79% match1 repo92024

MR-RawNet: Speaker verification system with multiple temporal resolutions for variable duration utterances using raw waveforms

78% match1 repo172024

Optimizing Contextual Speech Recognition Using Vector Quantization for Efficient Retrieval

78% match2024

Dynamic Cross Attention for Audio-Visual Person Verification

78% match2024

Cite this paper

@article{yip2023acanet,
  title  = {ACA-Net: Towards Lightweight Speaker Verification using Asymmetric Cross Attention},
  author = {Jia Qi Yip and Tuan Truong and Dianwen Ng and Chong Zhang and Yukun Ma and Trung Hieu Nguyen and Chongjia Ni and Shengkui Zhao and Eng Siong Chng and Bin Ma},
  year   = {2023},
  eprint = {2305.12121},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2305.12121v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2305.12121.svg)](https://codeofpaper.com/paper/2305.12121)

Discussion