Fact · Speech recognition
Graves wrote in 2012 that recurrent networks traditionally required a pre-defined alignment between input and output sequences, and introduced an end-to-end transduction system that removes that need.
Checked against its source on October 16, 2026
In the source's own words
“RNNs traditionally require a pre-defined alignment between the input and output sequences to perform transduction.”
Sequence Transduction with Recurrent Neural Networks, Alex Graves, arXiv (ICML 2012 Workshop on Representation Learning), 2012, Abstract.
Open the original- What it measures
- This is the idea behind the RNN transducer, the basis of most streaming speech recognizers on phones.
- What it does not show
- The paper tests only phoneme recognition on TIMIT; the speech-to-text transducers used today came later.
Source
Cited on
- Speech Recognition: What It Is and How Well It WorksTopic
- How Speech Recognition Works, From Sound to TextGuide
- Speech Recognition Models: How Neural ASR WorksGuide
- Real-Time Speech Recognition: Streaming and LatencyGuide
- History of Speech Recognition: From Audrey to WhisperGuide
- What role do hidden Markov models play in speech recognition?Question
- RNN transducer (RNN-T)Glossary