# On Low-Bit Quantization Errors in Speaker Verification: Diagnostic and Mitigation

arXiv: https://arxiv.org/abs/2606.08078 · HTML: https://mlsanity.com/paper/2606.08078/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 13 of its 21 references that are in the map.
- First posted on arXiv 6 Jun 2026.

## Builds on (most cited first, 13 of 13)

- [Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation](https://mlsanity.com/paper/1308.3432/) (2013)
- [VoxCeleb2: Deep Speaker Recognition](https://mlsanity.com/paper/1806.05622/) (2018)
- [VoxCeleb: a large-scale speaker identification dataset](https://mlsanity.com/paper/1706.08612/) (2017)
- [ECAPA-TDNN: Emphasized Channel Attention, Propagation and Aggregation in TDNN Based Speaker Verification](https://mlsanity.com/paper/2005.07143/) (2020)
- [MUSAN: A Music, Speech, and Noise Corpus](https://mlsanity.com/paper/1510.08484/) (2015)
- [BUT System Description to VoxCeleb Speaker Recognition Challenge 2019](https://mlsanity.com/paper/1910.12592/) (2019)
- [CN-CELEB: a challenging Chinese speaker recognition dataset](https://mlsanity.com/paper/1911.01799/) (2019)
- [Reshape Dimensions Network for Speaker Recognition](https://mlsanity.com/paper/2407.18223/) (2024)
- [ECAPA2: A Hybrid Neural Network Architecture and Training Strategy for Robust Speaker Embeddings](https://mlsanity.com/paper/2401.08342/) (2024)
- [Towards Lightweight Applications: Asymmetric Enroll-Verify Structure for Speaker Verification](https://mlsanity.com/paper/2110.04438/) (2021)
- [Studying squeeze-and-excitation used in CNN for speaker verification](https://mlsanity.com/paper/2109.05977/) (2021)
- [Model Compression for DNN-based Speaker Verification Using Weight Quantization](https://mlsanity.com/paper/2210.17326/) (2022)
- [Towards Lightweight Speaker Verification via Adaptive Neural Network Quantization](https://mlsanity.com/paper/2406.05359/) (2024)

## Abstract (arXiv)

Although low-bit quantization provides practical means to deploy speaker verification on resource-constrained devices, its effects on speaker verification performance remain poorly understood. In this paper, we study uniform K-means quantization-aware training of ResNet-36 and ResNet-200 through joint layer-wise and score-level analyses. Our layer-wise analysis highlights fragile components and shows that score degradation is not fully explained by weight distortion alone. We identify a clear knee point at 2 bits, with larger score drift and harmful decision flips concentrated near the FP32 threshold. Our score-level analysis reveals where and how score errors emerge under extreme quantization. Building on these findings, we propose a calibrated multi-precision cascade that resolves most trials at 2 bits and escalates only ambiguous cases, achieving performance close to FP32 while preserving the efficiency benefits of low-bit inference with substantially lower compute and memory costs.