EfficientTDNN: Efficient Architecture Search for Speaker Recognition
arXiv:2103.13581 · doi:10.1109/TASLP.2022.3182856
Abstract
Convolutional neural networks (CNNs), such as the time-delay neural network (TDNN), have shown their remarkable capability in learning speaker embedding. However, they meanwhile bring a huge computational cost in storage size, processing, and memory. Discovering the specialized CNN that meets a specific constraint requires a substantial effort of human experts. Compared with hand-designed approaches, neural architecture search (NAS) appears as a practical technique in automating the manual architecture design process and has attracted increasing interest in spoken language processing tasks such as speaker recognition. In this paper, we propose EfficientTDNN, an efficient architecture search framework consisting of a TDNN-based supernet and a TDNN-NAS algorithm. The proposed supernet introduces temporal convolution of different ranges of the receptive field and feature aggregation of various resolutions from different layers to TDNN. On top of it, the TDNN-NAS algorithm quickly searches for the desired TDNN architecture via weight-sharing subnets, which surprisingly reduces computation while handling the vast number of devices with various resources requirements. Experimental results on the VoxCeleb dataset show the proposed EfficientTDNN enables approximate architectures concerning depth, kernel, and width. Considering different computation constraints, it achieves a 2.20% equal error rate (EER) with 204M multiply-accumulate operations (MACs), 1.41% EER with 571M MACs as well as 0.94% EER with 1.45G MACs. Comprehensive investigations suggest that the trained supernet generalizes subnets not sampled during training and obtains a favorable trade-off between accuracy and efficiency.
arXiv comment13 pages, 12 figures, accepted to TASLP
References in corpus (17)
- Neural Architecture Search with Reinforcement Learning
- SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition
- Res2Net: A New Multi-scale Backbone Architecture
- VoxCeleb2: Deep Speaker Recognition
- Squeeze-and-Excitation Networks
- ECAPA-TDNN: Emphasized Channel Attention, Propagation and Aggregation in TDNN Based Speaker Verification
- DARTS: Differentiable Architecture Search
- EfficientNetV2: Smaller Models and Faster Training
- In defence of metric learning for speaker recognition
- Clova Baseline System for the VoxCeleb Speaker Recognition Challenge 2020
- Unified Hypersphere Embedding for Speaker Recognition
- VoxSRC 2019: The first VoxCeleb Speaker Recognition Challenge
- Semi-Supervised Neural Architecture Search
- Weight-Sharing Neural Architecture Search: A Battle to Shrink the Optimization Gap
- How to Train Your Super-Net: An Analysis of Training Heuristics in Weight-Sharing NAS
- Learning Autoencoders with Relational Regularization
- SpeechNAS: Towards Better Trade-off between Latency and Accuracy for Large-Scale Speaker Verification