Preprint

Listen, Attend and Spell

William Chan et al., arXiv, 2015

Read on October 17, 2026

What it is: The paper that introduced Listen, Attend and Spell (LAS), an attention-based encoder-decoder that writes characters from filter bank spectra without the independence assumption of CTC models. It reports word error rates on a subset of Google voice search.
How much weight it carries: Preprint from Google researchers, widely cited. Results are on a voice search subset, and the figures change between arXiv versions, so cite the version that matches the claim. Numbers from 2015 are not a measure of current systems.

Facts taken from this source

The LAS paper says its network produces character sequences without any independence assumptions between the characters, which it names as the key improvement over previous end-to-end CTC models.
“The network produces character sequences without making any independence assumptions between the characters. This is the key improvement of LAS over previous end-to-end CTC models.” (Abstract)
On a subset of Google's voice search task, the LAS model reached a word error rate of 14.1 percent without a dictionary or a language model.
“On a subset of the Google voice search task, LAS achieves a word error rate (WER) of 14.1% without a dictionary or a language model” (Abstract)