LLM-ForcedAligner: A Non-Autoregressive and Accurate LLM-Based Forced Aligner for Multilingual and Long-Form Speech

TL;DR

The paper introduces LLM-ForcedAligner, a novel non-autoregressive, multilingual speech alignment method using large language models reformulated as a slot-filling task, achieving high accuracy and speed in long-form speech scenarios.

Contribution

It presents a new reformulation of forced alignment as a slot-filling task with dynamic slot insertion, enabling non-autoregressive inference and improved multilingual performance.

Findings

Achieves 69-78% reduction in alignment shift compared to prior methods.

Supports arbitrary position alignment through dynamic slot insertion.

Operates efficiently on long-form and multilingual speech data.

Abstract

Forced alignment (FA) predicts start and end timestamps for words or characters in speech, but existing methods are language-specific and prone to cumulative temporal shifts. The multilingual speech understanding and long-sequence processing abilities of speech large language models (SLLMs) make them promising for FA in multilingual, crosslingual, and long-form speech settings. However, directly applying the next-token prediction paradigm of SLLMs to FA results in hallucinations and slow inference. To bridge the gap, we propose LLM-ForcedAligner, reformulating FA as a slot-filling paradigm: timestamps are treated as discrete indices, and special timestamp tokens are inserted as slots into the transcript. Conditioned on the speech embeddings and the transcript with slots, the SLLM directly predicts the time indices at slots. During training, causal attention masking with non-shifted input and label sequences allows each slot to predict its own timestamp index based on itself and preceding context, with loss computed only at slot positions. Dynamic slot insertion enables FA at arbitrary positions. Moreover, non-autoregressive inference is supported, avoiding hallucinations and improving speed. Experiments across multilingual, crosslingual, and long-form speech scenarios show that LLM-ForcedAligner achieves a 69%~78% relative reduction in accumulated averaging shift compared with prior methods. Checkpoint and inference code are available at https://github.com/QwenLM/Qwen3-ASR.