Preprint
Speech Recognition with Deep Recurrent Neural Networks
Alex Graves et al., arXiv (accepted for ICASSP 2013), 2013
Read the originalRead on October 17, 2026
- What it is
- A paper by Graves, Mohamed and Hinton showing that deep long short-term memory recurrent networks, trained end-to-end with Connectionist Temporal Classification, reached a strong phoneme recognition score on the TIMIT benchmark.
- How much weight it carries
- Preprint with a stated ICASSP 2013 acceptance and widely cited. The result is on TIMIT phoneme recognition, not on word error rate for full conversational speech, so it measures an earlier task than the modern benchmarks in this guide.
Facts taken from this source
- Graves, Mohamed and Hinton report that deep long short-term memory recurrent networks, trained end-to-end with suitable regularisation, reached a 17.7% test set error on the TIMIT phoneme recognition benchmark.
“When trained end-to-end with suitable regularisation, we find that deep Long Short-term Memory RNNs achieve a test set error of 17.7% on the TIMIT phoneme recognition benchmark” (Abstract)