🤖 AI Summary
Traditional speaker verification models (e.g., TDNN, ECAPA-TDNN) overly rely on long-term contextual information while neglecting fine-grained speaker characteristics, leading to limited discriminative power. To address this, we propose the Multi-granularity Feature Fusion TDNN (M-TDNN). Our method introduces three key innovations: (1) a 2D depthwise separable convolutional frontend to enhance local time-frequency modeling; (2) the first integration of phoneme-level feature pooling with TDNN, explicitly capturing phoneme-scale discriminative cues; and (3) a triple-domain fusion mechanism combining time-frequency, contextual, and fine-grained features. Evaluated on VoxCeleb1, M-TDNN achieves state-of-the-art performance—significantly reducing EER—while requiring fewer parameters and lower computational cost than standard TDNN and ECAPA-TDNN baselines.
📝 Abstract
In speaker verification, traditional models often emphasize modeling long-term contextual features to capture global speaker characteristics. However, this approach can neglect fine-grained voiceprint information, which contains highly discriminative features essential for robust speaker embeddings. This paper introduces a novel model architecture, termed MGFF-TDNN, based on multi-granularity feature fusion. The MGFF-TDNN leverages a two-dimensional depth-wise separable convolution module, enhanced with local feature modeling, as a front-end feature extractor to effectively capture time-frequency domain features. To achieve comprehensive multi-granularity feature fusion, we propose the M-TDNN structure, which integrates global contextual modeling with fine-grained feature extraction by combining time-delay neural networks and phoneme-level feature pooling. Experiments on the VoxCeleb dataset demonstrate that the MGFF-TDNN achieves outstanding performance in speaker verification while remaining efficient in terms of parameters and computational resources.