Connectionist temporal classification

TLDR

Sequence learning tasks often require predicting label sequences from noisy, unsegmented data, as in speech recognition, yet recurrent neural networks need pre‑segmented training data and post‑processing, limiting their use. The authors propose a novel training method that enables RNNs to learn label sequences directly from unsegmented input, eliminating the need for pre‑segmentation and post‑processing. This framework trains RNNs to map noisy, unsegmented signals to label sequences without requiring explicit segmentation during training or inference. Experiments on the TIMIT speech corpus show that the proposed method outperforms both a baseline HMM and a hybrid HMM‑RNN.

Abstract

Many real-world sequence learning tasks require the prediction of sequences of labels from noisy, unsegmented input data. In speech recognition, for example, an acoustic signal is transcribed into words or sub-word units. Recurrent neural networks (RNNs) are powerful sequence learners that would seem well suited to such tasks. However, because they require pre-segmented training data, and post-processing to transform their outputs into label sequences, their applicability has so far been limited. This paper presents a novel method for training RNNs to label unsegmented sequences directly, thereby solving both problems. An experiment on the TIMIT speech corpus demonstrates its advantages over both a baseline HMM and a hybrid HMM-RNN.