Research

Paper

AI LLM March 09, 2026

NLE: Non-autoregressive LLM-based ASR by Transcript Editing

Authors

Avihu Dekel, Samuel Thomas, Takashi Fukada, George Saon

Abstract

While autoregressive (AR) LLM-based ASR systems achieve strong accuracy, their sequential decoding limits parallelism and incurs high latency. We propose NLE, a non-autoregressive (NAR) approach that formulates speech recognition as conditional transcript editing, enabling fully parallel prediction. NLE extracts acoustic embeddings and an initial hypothesis from a pretrained speech encoder, then refines the hypothesis using a bidirectional LLM editor trained with a latent alignment objective. An interleaved padding strategy exploits the identity mapping bias of Transformers, allowing the model to focus on corrections rather than full reconstruction. On the Open ASR leaderboard, NLE++ achieves 5.67% average WER with an RTFx (inverse real-time factor) of 1630. In single-utterance scenarios, NLE achieves 27x speedup over the AR baseline, making it suitable for real-time applications.

Metadata

arXiv ID: 2603.08397
Provider: ARXIV
Primary Category: eess.AS
Published: 2026-03-09
Fetched: 2026-03-10 05:43

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.08397v1</id>\n    <title>NLE: Non-autoregressive LLM-based ASR by Transcript Editing</title>\n    <updated>2026-03-09T13:56:52Z</updated>\n    <link href='https://arxiv.org/abs/2603.08397v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.08397v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>While autoregressive (AR) LLM-based ASR systems achieve strong accuracy, their sequential decoding limits parallelism and incurs high latency. We propose NLE, a non-autoregressive (NAR) approach that formulates speech recognition as conditional transcript editing, enabling fully parallel prediction. NLE extracts acoustic embeddings and an initial hypothesis from a pretrained speech encoder, then refines the hypothesis using a bidirectional LLM editor trained with a latent alignment objective. An interleaved padding strategy exploits the identity mapping bias of Transformers, allowing the model to focus on corrections rather than full reconstruction. On the Open ASR leaderboard, NLE++ achieves 5.67% average WER with an RTFx (inverse real-time factor) of 1630. In single-utterance scenarios, NLE achieves 27x speedup over the AR baseline, making it suitable for real-time applications.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='eess.AS'/>\n    <published>2026-03-09T13:56:52Z</published>\n    <arxiv:comment>Preprint</arxiv:comment>\n    <arxiv:primary_category term='eess.AS'/>\n    <author>\n      <name>Avihu Dekel</name>\n    </author>\n    <author>\n      <name>Samuel Thomas</name>\n    </author>\n    <author>\n      <name>Takashi Fukada</name>\n    </author>\n    <author>\n      <name>George Saon</name>\n    </author>\n  </entry>"
}