Don McAllaster

Learn More
We present a study of data simulated using acoustic models trained on Switchboard data, and then recognized using various Switchboard-trained acoustic models. When we recognize real Switchboard conversations, simple development models give a word error rate (WER) of about 47 percent. If instead we simulate the speech data using word transcriptions of the(More)
This paper describes recent changes in Dragon’s speech recognition system which have markedly improved performance on conversational telephone speech. Key changes include: the conversion to modified PLP-based cepstra from mel-cepstra; the replacement of our usual IMELDA transformation by a new transform using “semi-tied covariance”; a new multi-pass(More)
In this paper we revisit a topic identification test on the Switchboard Corpus first reported at ICASSP’93. Dragon’s approach to topic ID uses a large vocabulary continuous speech recognizer as a front-end to transcribe the speech and then scores the transcripts using a set of topic-specific language models. Our recognition of conversational telephone(More)
We present a study of data simulated using acoustic models trained on Switchboard data, and then recognized using various Switchboard-trained acoustic models. The Switchboard-trained models yield word error rates of about 47 percent, on real Switchboard conversations. When data is simulated using the acoustic models, but in a way that insures that the(More)
In this paper, we introduce a new conversational speech task – recognizing call-center speech – using data collected from Dragon’s own technical support line. We compare performance of models trained from conversational telephone speech (the Switchboard corpus) and models trained from predominantly read, microphone speech, and report on a series of(More)
We continue our study of the use of fabricated data in the investigation of speech recognition algorithms. After reviewing the basic data generation algorithm and some earlier results involving the recognition of fabricated conversational speech data, we go on to describe some new and intriguing experiments concerning on the one hand, training acoustic(More)
We report results of large vocabulary continuous speech recognition (LVCSR) experiments, conducted using speech data read over cellular and landline phones. Specifically, we compare (using stereo recordings) the speaker-independent and speakeradapted recognition word error rates (WERs) measured over cellular and landline networks, with those measured using(More)
We present a study of a speaker verification system for telephone data based on large-vocabulary speech recognition. After describing the recognition engine, we give details of the verification algorithm and draw comparisons with other systems. The system has been tested on a test set taken from the Switchboard corpus of conversational telephone speech, and(More)
  • 1