AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Multi-granularity Interactive Attention Framework for Residual Hierarchical Pronunciation Assessment

Hong Han, Hao-Chen Pei, Zhao-Zheng Nie, Xin Luo, Xin-Shun Xu

AAAI 2026 · Architectures and efficiency › Attention and transformers · approach: Long-context KV and memory attention

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add interactive attention heads at all granularities (ablation)HIA · speechocean762word stress PCC: 0.335 -> 0.436Improved
add interactive attention heads at all granularities (ablation)HIA · speechocean762utterance total PCC: 0.748 -> 0.764Improved
add residual connection from encoder embeddings (ablation)HIA · speechocean762word stress PCC: 0.382 -> 0.436Improved
add hierarchical score passing between granularities (ablation)HIA · speechocean762word stress PCC: 0.374 -> 0.436Improved
add one convolutional layer per granularity (ablation)HIA · speechocean762word total PCC: 0.601 -> 0.628Improved
increase convolutional layers from 1 to 2 or 3 (ablation)HIA · speechocean762utterance total PCC: 0.764 -> 0.759 (2 layers), 0.755 (3 layers)Hurt
increase embedding size from 48 to 96 (ablation)HIA · speechocean762word total PCC: 0.628 -> 0.611Hurt
reduce embedding size from 48 to 24 (ablation)HIA · speechocean762utterance total PCC: 0.764 -> 0.752Hurt
use multiple attention heads instead of one (ablation)HIA · speechocean762utterance total PCC: 0.764 -> 0.759 (2 heads), 0.751 (4 heads)Hurt
HIA vs prior multi-granularity SOTA GradformerHIA · speechocean762phoneme MSE: 0.079 -> 0.076Improved
HIA vs single-granularity LAS on utterance totalHIA · speechocean762utterance total PCC: 0.766 (LAS) vs 0.764Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy