# Recurrent Neural Networks with Stochastic Layers for Acoustic Novelty Detection

## TL;DR

This paper introduces a novel recurrent neural network with stochastic layers for acoustic novelty detection, leveraging uncertainty modeling to improve detection performance in an unsupervised, end-to-end manner.

## Contribution

It adapts stochastic RNNs for acoustic novelty detection, enabling explicit probability calculations and improved robustness over existing methods.

## Key findings

- Outperforms state-of-the-art detectors on benchmark datasets
- Robust and highly unsupervised with minimal preprocessing
- Requires less hyperparameter tuning

## Abstract

In this paper, we adapt Recurrent Neural Networks with Stochastic Layers, which are the state-of-the-art for generating text, music and speech, to the problem of acoustic novelty detection. By integrating uncertainty into the hidden states, this type of network is able to learn the distribution of complex sequences. Because the learned distribution can be calculated explicitly in terms of probability, we can evaluate how likely an observation is then detect low-probability events as novel. The model is robust, highly unsupervised, end-to-end and requires minimum preprocessing, feature engineering or hyperparameter tuning. An experiment on a benchmark dataset shows that our model outperforms the state-of-the-art acoustic novelty detectors.