Large Vocabulary Continuous Speech Recognition System Research Articles

Automatic generation of subword units for speech recognition systems

Large vocabulary continuous speech recognition (LVCSR) systems traditionally represent words in terms of smaller subword units. Both during training and during recognition, they require a mapping table, called the dictionary, which maps words into sequences of these subword units. The performance of the LVCSR system depends critically on the definition of the subword units and the accuracy of the dictionary. In current LVCSR systems, both these components are manually designed. While manually designed subword units generalize well, they may not be the optimal units of classification for the specific task or environment for which an LVCSR system is trained. Moreover, when human expertise is not available, it may not be possible to design good subword units manually. There is clearly a need for data-driven design of these LVCSR components. In this paper, we present a complete probabilistic formulation for the automatic design of subword units and dictionary, given only the acoustic data and their transcriptions. The proposed framework permits easy incorporation of external sources of information, such as the spellings of words in terms of a nonideographic script.

Speech Recognition Using Acoustic Similarity-Based Primitives

This paper proposes an algorithm that automatically acquires a new recognition primitive by splitting the training sample so that the likeness of the whole training sample to the model is maximized. The primitive obtained by the proposed method is a primitive integrating several time-continuous phonemes. It is called an acoustic similarity-based primitive (ASP). The algorithm proposed in this paper performs in parallel ASP acquisition and ASP modeling by HMnet. The two are simultaneously optimized. In phoneme recognition experiments for six specified speakers, the recognition rate was improved by approximately 3.5% on the average, compared to the conventional phoneme HMnet, by increasing the number of candidates by approximately 14.6% on the average. A method is also proposed in which an ASP is built in a large-vocabulary continuous speech recognition system. In a recognition experiment with eight specified speakers, the accuracy of word recognition was improved by approximately 2.6%, compared to the phoneme HMnet. © 2001 Scripta Technica, Syst Comp Jpn, 33(1): 8–17, 2002

Japanese large-vocabulary continuous-speech recognition using a newspaper corpus and broadcast news

This paper describes large-vocabulary continuous-speech recognition (LVCSR) of Japanese newspaper speech read aloud and Japanese broadcast-news speech. It describes the first Japanese LVCSR experiments using morpheme-based statistical language models. The statistical language models were trained using a large text corpus constructed from several years of newspaper texts and our LVCSR system was evaluated using recorded newspaper speech read by 10 male speakers. It is difficult to train statistical n-gram language models for Japanese because Japanese sentences are written without spaces between words. This difficulty was overcome by segmenting sentences into words with a morphological analyzer and then training the n-gram language models using those words. The LVCSR system was constructed with the language models trained using the newspaper articles, and the acoustic models, which were phoneme hidden Markov models (HMMs) trained using 20 h of speech. The results for recognition of read newspaper speech with a 7k vocabulary were comparable to those for other languages. For the automatic transcription of broadcast-news speech with our LVCSR system, the language models had 20k word vocabularies and were trained using broadcast-news manuscripts. These models achieved better performance than the language models trained using newspaper texts. Our experiments indicate that LVCSR for Japanese works in much the same way as LVCSR for European languages.

A VLSI grammar processing subsystem for a real-time large-vocabulary continuous speech recognition system

The architecture and custom implementation of a grammar processing subsystem for a real-time large-vocabulary continuous speech recognition system using hidden Markov models are described. The prototype has 3000, words and larger vocabularies and higher throughput can be obtained by adding more grammar processors. This subsystem contains two custom VLSI chips that perform the evaluation of starting word probabilities associated with the across-word transitions in the hidden-Markov-model (HMM)-based speech recognition system. This system has a maximum computation rate of 200 MOPS and an I/O bandwidth of 265 MB/s. All circuits were silicon compiled and were working on first silicon.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Integrated circuits for a real-time large-vocabulary continuous speech recognition system

The architecture and implementation of a word processing subsystem for a real-time speech recognition system using hidden Markov models are described. The bottleneck of this system, which is the acquisition of data, is demonstrated, and an architecture that speeds up this bottleneck using on-chip dual-ported cache memories is presented. The architecture is described in a textual form, and the layout data were completely automatically generated. The chips have been fabricated through MOSIS using a 2- mu m CMOS n-well technology. The functionality of the processors was successfully tested using the scan-path test methodology. The clock rate for the scan-path test was 5 MHz to guarantee proper operation of the circuits for this clock rate. All the processors were first time working silicon.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>