# Segmental Recurrent Neural Networks for End-to-end Speech Recognition

arXiv: https://arxiv.org/abs/1603.00223 · HTML: https://mlsanity.com/paper/1603.00223/

## Key facts

- Cited by 7 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 1 Mar 2016.

## Built on by (most cited first, 7 of 7)

- [Streaming Multi-talker Speech Recognition with Joint Speaker Identification](https://mlsanity.com/paper/2104.02109/) (2021)
- [A study of latent monotonic attention variants](https://mlsanity.com/paper/2103.16710/) (2021)
- [Normalizing Flow based Hidden Markov Models for Classification of Speech Phones with Explainability](https://mlsanity.com/paper/2107.00730/) (2021)
- [Monotonic segmental attention for automatic speech recognition](https://mlsanity.com/paper/2210.14742/) (2022)
- [Joint Span Segmentation and Rhetorical Role Labeling with Data Augmentation for Legal Documents](https://mlsanity.com/paper/2302.06448/) (2023)
- [Transcription free filler word detection with Neural semi-CRFs](https://mlsanity.com/paper/2303.06475/) (2023)
- [SynJax: Structured Probability Distributions for JAX](https://mlsanity.com/paper/2308.03291/) (2023)

## Abstract (arXiv)

We study the segmental recurrent neural network for end-to-end acoustic modelling. This model connects the segmental conditional random field (CRF) with a recurrent neural network (RNN) used for feature extraction. Compared to most previous CRF-based acoustic models, it does not rely on an external system to provide features or segmentation boundaries. Instead, this model marginalises out all the possible segmentations, and features are extracted from the RNN trained together with the segmental CRF. In essence, this model is self-contained and can be trained end-to-end. In this paper, we discuss practical training and decoding issues as well as the method to speed up the training in the context of speech recognition. We performed experiments on the TIMIT dataset. We achieved 17.3 phone error rate (PER) from the first-pass decoding --- the best reported result using CRFs, despite the fact that we only used a zeroth-order CRF and without using any language model.