Peer-reviewed research
Connectionist Temporal Classification: Labelling Unsegmented Sequence Data with Recurrent Neural Networks
Alex Graves et al., Proceedings of the 23rd International Conference on Machine Learning (ICML 2006), pp. 369-376, 2006 · DOI 10.1145/1143844.1143891
Read the originalRead on October 17, 2026
- What it is
- The paper that introduced connectionist temporal classification (CTC), a way to train a recurrent neural network to output a label sequence from audio without a frame-by-frame alignment. Its motivating example is speech recognition, and its experiments use the TIMIT speech corpus.
- How much weight it carries
- Peer-reviewed at ICML and widely cited. The experiments are on TIMIT, a small phone-recognition benchmark from 1990s-era data, so the gains do not show how the method performs on today's large word-level tasks.
Facts taken from this source
- Graves and colleagues wrote that recurrent networks needed pre-segmented training data and post-processing to produce label sequences, and presented a method that trains them to label unsegmented sequences directly.
“because they require pre-segmented training data, and post-processing to transform their outputs into label sequences, their applicability has so far been limited” (Abstract)
Pages that rest on this source
- How Speech Recognition Works, From Sound to TextGuide
- Speech Recognition Models: How Neural ASR WorksGuide
- History of Speech Recognition: From Audrey to WhisperGuide
- Is speech recognition AI?Question
- What are the most important speech recognition research papers?Question
- Who invented speech recognition?Question
- Connectionist temporal classification (CTC)Glossary