Unsupervised Training on Large Amounts of Broadcast News Data

Abstract

This paper presents our recent effort that aims at improving our Arabic Broadcast News (BN) recognition system by using thousands of hours of un-transcribed Arabic audio in the way of unsupervised training. Unsupervised training is first carried out on the 1,900-hour English Topic Detection and Tracking (TDT) data and is compared with the lightly-supervised training method that we have used for the DARPA EARS evaluations. The comparison shows that unsupervised training produces a 21.7% relative reduction in word error rate (WER), which is comparable to the gain obtained with light supervision methods. The same unsupervised training strategy carried out on a similar amount of Arabic BN data produces an 11.6% relative gain. The gain, though considerable, is substantially smaller than what is observed on the English data. Our initial work towards understanding the reasons for this difference is also described.

Deep neural network features and semi-supervised training for low resource speech recognition

“…Semi-supervised training has been effectively used to train acoustic models in several languages and conditions [32,33,34,35,36]. This section discusses the application of these approaches to low-resource settings.…”

A summary of the 2012 JHU CLSP workshop on zero resource speech technologies and models of early language acquisition

“…In low resource scenarios, we seek multi-lingual and semi-supervised methods to leverage more easily acquired high-resource or untranscribed speech to improve our ASR performance with minimal cost. Two avenues were explored in the workshop: (i) a multi-lingual corpus was used to train a data-driven, language-invariant front-end for low-resource recognition; and (ii) untranscribed speech audio was automatically transcribed and used to augment to the labeled data for training, a procedure known as self-training [33,34]. For (i), discriminative deep neural network (DNN) pre-training [35] was performed on a multilingual corpus consisting of 31 hours of German/Spanish and only a single hour of English.…”

Combination of data borrowing strategies for low-resource LVCSR

“…In addition unsupervised [9] or lightlysupervised [10] training is another type of popular strategy which could enlarge the size of target language data quickly and cheaply [11] [12]. The multilingual or cross-lingual approaches borrow data from the V2 (transcribed non-target language data), and the unsupervised training usually develop technologies to borrow data from the untranscribed target language data, i.e.…”