Robust cross-domain disfluency detection with pattern match networks
This addresses disfluency detection for speech processing, offering a robust cross-domain solution, though it is incremental as it builds on existing pattern match methods.
The paper tackled disfluency detection across speech genres by introducing a pattern match neural network that uses neighbor similarity scores, eliminating feature engineering; it matched hand-engineered features in-domain and achieved superior cross-domain performance.
In this paper we introduce a novel pattern match neural network architecture that uses neighbor similarity scores as features, eliminating the need for feature engineering in a disfluency detection task. We evaluate the approach in disfluency detection for four different speech genres, showing that the approach is as effective as hand-engineered pattern match features when used on in-domain data and achieves superior performance in cross-domain scenarios.