G-IFT: A Gated Linear Unit adapter with Iterative Fine-Tuning for Low-Resource Children's Speaker Verification

2025-08-11

AI Summary

Limited child speech data severely degrades the transferability of adult-pretrained speaker verification models, primarily due to acoustic distribution shift and insufficient fine-tuning. To address this, we propose a cross-domain knowledge transfer framework featuring gated linear unit (GLU)-based adapters and staged iterative fine-tuning, enabling efficient, lightweight adaptation of mainstream embedding backbones—including ECAPA-TDNN, ResNet, and x-vector. The GLU adapter explicitly models child-specific acoustic representations, while joint optimization of the pretrained backbone, adapter, and classifier enhances generalization under low-resource conditions. Experiments on the OGI and MyST child speech datasets demonstrate consistent improvements: equal error rates (EER) decrease by 18.7%–26.3% over strong baselines. The framework exhibits both effectiveness and architecture-agnostic transferability across diverse speaker embedding models.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Natural Language Processing: Speech
• Cognitive Modeling & Cognitive Systems: Adaptive Behavior

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Large pretrained models with web data
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Speaker Verification (SV) systems trained on adults speech often underperform on children's SV due to the acoustic mismatch, and limited children speech data makes fine-tuning not very effective. In this paper, we propose an innovative framework, a Gated Linear Unit adapter with Iterative Fine-Tuning (G-IFT), to enhance knowledge transfer efficiency between the high-resource adults speech domain and the low-resource children's speech domain. In this framework, a Gated Linear Unit adapter is first inserted between the pre-trained speaker embedding model and the classifier. Then the classifier, adapter, and pre-trained speaker embedding model are optimized sequentially in an iterative way. This framework is agnostic to the type of the underlying architecture of the SV system. Our experiments on ECAPA-TDNN, ResNet, and X-vector architectures using the OGI and MyST datasets demonstrate that the G-IFT framework yields consistent reductions in Equal Error Rates compared to baseline methods.

Problem

• Improve children's speaker verification with limited data
• Address acoustic mismatch between adult and child speech
• Enhance knowledge transfer from adult to child speech domains

Innovation

• Gated Linear Unit adapter for knowledge transfer
• Iterative fine-tuning of model components
• Agnostic to underlying speaker verification architecture