Assessing the Energy and Carbon Emissions of Neural Speaker Verification Model in Training and Inference

2026-06-06

AI Summary

This study addresses the lack of systematic evaluation of energy consumption and carbon emissions associated with deep neural networks in speaker verification. For the first time, it employs node-level power sensors to empirically measure the energy usage and carbon footprint of ResNet models of varying scales during both training and inference on the VoxCeleb2 dataset. The work systematically analyzes the trade-offs between model depth, width, computational stage distribution, accuracy, and environmental impact. Findings reveal that medium-scale architectures such as ResNet-50, particularly those with stage-concentrated designs, achieve an optimal balance between performance and energy efficiency. The study further uncovers diminishing marginal returns with model scaling and proposes design principles for energy-efficient, environmentally sustainable speaker verification systems.

Abstract

Deep-learning speaker verification (SV) increasingly relies on deep neural network backbones, whose environmental impact remains largely undocumented. In this paper, we conduct an evaluation of ResNet architectures trained on VoxCeleb2, varying depth, channel width, and stage distribution, and measure energy consumption and carbon footprint using node-level sensors. Results show a clear point of diminishing returns: deeper or wider models bring only marginal accuracy gains while energy consumption grows steeply. In contrast, mid-sized networks such as ResNet-50 and stage-concentrated variants achieve favorable trade-offs between performance and environmental impact. These findings provide actionable guidelines for designing energy-efficient SV systems.

Problem

• speaker verification
• energy consumption
• carbon emissions
• deep neural networks
• environmental impact

Innovation

• energy efficiency
• carbon footprint
• speaker verification
• ResNet architecture
• environmental impact