Hybrid Autoregressive Transducer (HAT)

By Ehsan Variani, David Rybach, Cyril Allauzen, Michael D. Riley. 2020

174 citations recorded by the source corpus.

This paper proposes and evaluates the hybrid autoregressive transducer (HAT) model, a time-synchronous encoder-decoder model that preserves the modularity of conventional automatic speech recognition systems. The HAT model provides a way to measure the quality of the internal language model that can be used to decide whether inference with an external language model is beneficial or not. We evaluate our proposed model on a large-scale voice search task. Our experiments show significant improvements in WER compared to the state-of-the-art approaches1.

Subfields: Model Architectures, Speech & Audio

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Model Architectures

In Speech & Audio

Citations in the corpus

Multitask Training with Text Data for End-to-End Speech Recognition (2020, methodology): “Compared with JATD, a more closely related area of MUTE is the subtraction of internal LMs for end-to-end ASR models [30, 31].”
End-to-End Speech Recognition: A Survey (2023, background): “Hybrid autoregressive transducer (HAT) [34] extends RNN-T so that the model becomes the internal LM when the encoder output is eliminated, i.”
Recent Advances in End-to-End Automatic Speech Recognition (2021, methodology): “Another similar model is the hybrid autoregressive transducer (HAT) model [285] designed to improve the RNN-T model.”
Injecting Text in Self-Supervised Speech Pretraining (2021, unknown): “…such as “Shallow Fusion” [30] that interpolate scores from the end-to-end model and an external language model (LM) and approaches that jointly train end to end models and LMs, such as “Cold Fusion” [31], “Deep Fusion” [30], “Component Fusion” [32] and Hybrid Autoregressive Transducers (HAT)[33].”
Less is More: Improved RNN-T Decoding Using Limited Label Context and Path Merging (2020, background): “[19] find that the hybrid autoregressive transducer (HAT; RNN-T with an ‘internal language model (LM)’), trained to output phonemes and decoded with a separate lexicon and grammar achieves similar performance when context is limited to two previous phoneme labels on a large scale task.”