# Graph-Based Phonetic Error Correction of Noisy ASR

arXiv: https://arxiv.org/abs/2606.24889 · HTML: https://mlsanity.com/paper/2606.24889/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 9 of its 16 references that are in the map.
- First posted on arXiv 29 Apr 2026.

## Builds on (most cited first, 9 of 9)

- [Inductive Representation Learning on Large Graphs](https://mlsanity.com/paper/1706.02216/) (2017)
- [Unsupervised Cross-lingual Representation Learning at Scale](https://mlsanity.com/paper/1911.02116/) (2019)
- [Learning ASR-Robust Contextualized Embeddings for Spoken Language Understanding](https://mlsanity.com/paper/1909.10861/) (2019)
- [WER we are and WER we think we are](https://mlsanity.com/paper/2010.03432/) (2020)
- [SeMaScore : a new evaluation metric for automatic speech recognition tasks](https://mlsanity.com/paper/2401.07506/) (2024)
- [Loquacious Set: 25,000 Hours of Transcribed and Diverse English Speech Recognition Data for Research and Commercial Use](https://mlsanity.com/paper/2505.21578/) (2025)
- [Fewer Hallucinations, More Verification: A Three-Stage LLM-Based Framework for ASR Error Correction](https://mlsanity.com/paper/2505.24347/) (2025)
- [Integrated Semantic and Phonetic Post-correction for Chinese Speech Recognition](https://mlsanity.com/paper/2111.08400/) (2021)
- [DoCIA: An Online Document-Level Context Incorporation Agent for Speech Translation](https://mlsanity.com/paper/2504.05122/) (2025)

## Abstract (arXiv)

Automatic speech recognition (ASR) systems, despite low overall word error rates, produce residual lexical errors that disproportionately affect semantically critical tokens such as named entities, negations, and sentiment-bearing words. These errors are often structured, arising from phonetic similarity rather than random noise, making naive token-level correction insufficient. We propose a structured ASR correction framework, that we call G-SPIN, that combines phonetic graph modeling with contextual language understanding. A graph neural network (GNN) first constructs acoustically plausible candidate neighborhoods for flagged tokens, explicitly restricting the correction search space to phonetic alternatives. A masked language model (MLM) then provides local contextual scoring, and an instruction-tuned large language model (LLM) performs final context-aware re-ranking over this compact candidate set. By decoupling structured phonetic reasoning from contextual semantic selection, our method avoids unconstrained generation while improving correction accuracy. The framework is lightweight, modular, and operates entirely at inference time.