Fact · Speech recognition
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.
Checked against its source on October 17, 2026
In the source's own words
“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.”
Listen, Attend and Spell, William Chan et al., arXiv, 2015, Abstract.
Open the original- What it measures
- CTC scores each output label on its own, given the audio. An attention decoder conditions each character on the characters already written, which is what the authors mean by no independence assumption.
- What it does not show
- A claim about the design of one 2015 model, as its authors describe it. Later CTC and transducer models added context in other ways, so the gap is architectural, not a fixed limit of CTC.