Commercial Speech Recognizers Research Articles

Noise Reduction in Industry Based on Virtual Instrumentation

This paper discusses the reduction of background noise in an industrial environment to extend human-machine-interaction. In the Industry 4.0 era, the mass development of voice control (speech recognition) in various industrial applications is possible, especially as related to augmented reality (such as hands-free control via voice commands). As Industry 4.0 relies heavily on radiofrequency technologies, some brief insight into this problem is provided, including the Internet of things (IoT) and 5G deployment. This study was carried out in cooperation with the industrial partner Brose CZ spol. s.r.o., where sound recordings were made to produce a dataset. The experimental environment comprised three workplaces with background noise above 100 dB, consisting of a laser/magnetic welder and a press. A virtual device was developed from a given dataset in order to test selected commands from a commercial speech recognizer from Microsoft. We tested a hybrid algorithm for noise reduction and its impact on voice command recognition efficiency. Using virtual devices, the study was carried out on large speakers with 20 participants (10 men and 10 women). The experiments included a large number of repetitions (100 times for each command under different noise conditions). Statistical results confirmed the efficiency of the tested algorithms. Laser welding environment efficiency was 27% before applied filtering, 76% using the least mean square (LMS) algorithm, and 79% using LMS + independent component analysis (ICA). Magnetic welding environment efficiency was 24% before applied filtering, 70% with LMS, and 75% with LMS + ICA. Press workplace environment efficiency showed no success before applied filtering, was 52% with LMS, and was 54% with LMS + ICA.

Web Services Based Hybrid Recognizer of Lithuanian Voice Commands

This paper presents the recently developed medical-pharmaceutical informative system with voice user interface. This is the first computerized system oriented towards healthcare services and industry where Lithuanian voice commands are used as a primary mean for control. Another essential property of the developed system is its hybrid nature: two different recognizers - an adapted commercial Spanish speech recognizer available from Microsoft and a locally developed HMM speech recognizer based on Lithuanian acoustic models – are operating in parallel. The recognition hypotheses produced by those recognizers are joined together using logical rules obtained using decision rules induction algorithms such as Ripper. All these measures and approaches allowed achieve very high speaker independent voice commands recognition accuracy acceptable for the system implementation in practice. The best achieved recognition was 98.9 % for 1000 Lithuanian voice commands. The paper presents optimization issues related with the development of the system. DOI: http://dx.doi.org/10.5755/j01.eee.20.9.8713

Comparative Analysis of Adapted Foreign Language and Native Lithuanian Speech Recognizers for Voice User Interface

Paper presents research results obtained when building a speaker independent hybrid speech recognizer. This recognizer will be integrated as a phrase recognizer in a medical-pharmaceutical information system. The hybrid speech recognizer consists of two recognition components: an adapted commercial Microsoft Spanish speech recognizer and a locally developed hidden Markov models based recognizer implementing Lithuanian acoustic models. Efficiency of both recognition components was evaluated on multiple speaker independent speech recognition tasks. The average accuracy of Lithuanian recognizer was higher reaching 0.6% phrase error rate for user requests in medical-pharmaceutical domain. The adapted commercial Spanish speech recognizer showed the ability to improve the accuracy of Lithuanian recognizer in the worst recognition scenarios. These results proved the hypothesis formulated when proposing the basic idea of hybrid recognition approach: recognition errors from different recognizers built using various techniques are not strongly correlated. This fact could be exploited for improved overall speech recognition accuracy. DOI: http://dx.doi.org/10.5755/j01.eee.19.7.5171

A hybrid noise suppression filter for accuracy enhancement of commercial speech recognizers in varying noisy conditions

• Commercial speech recognizers open speech control applications. • They are likely to be produced error when background noise surrounds. • Potential dangers are created if recognition errors exist. • We proposed a noise suppression filter to enhance the recognition accuracy. • It intends to decrease the recognition errors under noisy environment. Commercial speech recognizers have made possible many speech control applications such as wheelchair, tone-phone, multifunctional robotic arms and remote controls, for the disabled and paraplegic. However, they have a limitation in common in that recognition errors are likely to be produced when background noise surrounds the spoken command, thereby creating potential dangers for the disabled if recognition errors exist in the control systems. In this paper, a hybrid noise suppression filter is proposed to interface with the commercial speech recognizers in order to enhance the recognition accuracy under variant noisy conditions. It intends to decrease the recognition errors when the commercial speech recognizers are working under a noisy environment. It is based on a sigmoid function which can effectively enhance noisy speech using simple computational operations, while a robust estimator based on an adaptive-network-based fuzzy inference system is used to determine the appropriate operational parameters for the sigmoid function in order to produce effective speech enhancement under variant noisy conditions. The proposed hybrid noise suppression filter has the following advantages for commercial speech recognizers: (i) it is not possible to tune the inbuilt parameters on the commercial speech recognizers in order to obtain better accuracy; (ii) existing noise suppression filters are too complicated to be implemented for real-time speech recognition; and (iii) existing sigmoid function based filters can operate only in a single-noisy condition, but not under varying noisy conditions. The performance of the hybrid noise suppression filter was evaluated by interfacing it with a commercial speech recognizer, commonly used in electronic products. Experimental results show that improvement in terms of recognition accuracy and computational time can be achieved by the hybrid noise suppression filter when the commercial recognizer is working under various noisy environments in factories.

Enhancement of Speech Recognitions for Control Automation Using an Intelligent Particle Swarm Optimization

For over two decades, speech control mechanisms have been widely applied in manufacturing systems such as factory automation, warehouse automation, and industrial robotic control for over two decades. To implement speech controls, a commercial speech recognizer is used as the interface between users and the automation system. However, users' commands are often contaminated by environmental noise which degrades the performance of speech recognition for controlling automation systems. This paper presents a multichannel signal enhancement methodology to improve the performance of commercial speech recognizers. The proposed methodology aims to optimize speech recognition accuracy of a commercial speech recognizer in a noisy environment based on a beamformer, which is developed by an intelligent particle swarm optimization. It overcomes the limitation of the existing signal enhancement approaches whereby the parameters inside commercial speech recognizers are required to be tuned, which is impossible in a real-world situation. Also, it overcomes the limitation of the existing optimization algorithm including gradient descent methods, genetic algorithms and classical particle swarm optimization that are unlikely to develop optimal beamformers for maximizing speech recognition accuracy. The performance of the proposed methodology was evaluated by developing beamformers for a commercial speech recognizer, which was implemented on warehouse automation. Results indicate a significant improvement regarding speech recognition accuracy.

Recognition of voice commands using adaptation of foreign language speech recognizer via selection of phonetic transcriptions

Abstract In recent years various commercial speech recognizers have become available. These recognizers provide the possibility to develop applications incorporating various speech recognition techniques easily and quickly. All of these commercial recognizers are typically targeted to widely spoken languages having large market potential; however, it may be possible to adapt available commercial recognizers for use in environments where less widely spoken languages are used. Since most commercial recognition engines are closed systems the single avenue for the adaptation is to try set ways for the selection of proper phonetic transcription methods between the two languages. This paper deals with the methods to find the phonetic transcriptions for Lithuanian voice commands to be recognized using English speech engines. The experimental evaluation showed that it is possible to find phonetic transcriptions that will enable the recognition of Lithuanian voice commands with recognition accuracy of over 90%.

Evaluation of speech recognizers for speech training applications

The use of speech recognition technology for speech training represents an important and potentially very large application of speech technology. However, speech training places unique demands on recognizer performance that have not been well-characterized. In this research, a database and testing procedures were developed to evaluate two facets of recognizer performance integral to speech training: utterance identification and speech quality assessment. Using these materials, three commercial speech recognizers that employ different types of recognition algorithms were evaluated. In general, the recognizer, based on hidden Markov models (HMM's), provided better identification scores for normal and disordered speech than the two template-based recognizers. A recognizer's identification performance on normal speech often predicted its identification performance on disordered speech. For each recognizer, analysis using phonological features revealed classes of speech sounds that are poorly discriminated. Procedures were developed to provide human ratings of the quality of disordered speech for comparison to recognizer performance. Recognizers were compared to speech-language pathologists with respect to the ability to judge speech quality. In contrast, with identification performance, the two speech recognizers based on template comparisons provided better measures of speech quality than the HMM-based recognizer.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Speech input for dysarthric users

One practical aim of this research is to determine how best to use speech recognition techniques for augmenting the communication abilities of dysarthric speakers. As a first step toward this goal, the following kinds of analyses and tests have been performed on words produced by several dysarthric speakers: a closed‐set intelligibility test based on Kent et al. [J. Speech Hear. Disord. 54, 482–499 (1989)]; an open intelligibility test; critical listening and transcription; acoustic analysis of selected utterances; and an evaluation of the recognition of words by a commercial speech recognizer. The data from one speaker have been examined in detail. The analysis and testing have led to several conclusions concerning the control of the articulators for this speaker: production of obstruent consonants was a particular problem (only 30% of syllable‐initial obstruents were produced with no error), whereas sonorant consonants were less of a problem (70% correct). Of the obstruent errors, most were voicing errors, but place errors for alveolars (particularly fricatives) were also high, and these consonants were produced inconsistently, as inferred from acoustic analysis and from low scores from the recognizer for words with these consonants. In comparison, vowel errors were less prevalent. Implications for the use of a speech recognizer for augmenting this speaker’s communication abilities are discussed.

Temporal invariance for word identification

Attempts were made to learn how well timing alone could be used to differentiate words from each other when speakers and tempo vary. In addition, we wanted to determine if this information is being missed by some commercial speech recognizers. A set of 12 phrases and words was selected describable in terms of the same sequence of acoustic intervals but differing in stress, consonant voicing, vowel tensity, etc.: accuse, Otis, advise, it froze, outhouse, appease, abbeys, etc.). The lists were recorded at two tempos by four speakers and prominent intervals were measured. First, discriminant analysis was employed on the measurements to identify the words. Second, the recordings were used with a commercial isolated‐word speaker‐dependent recognizer for training and testing. Many simple experiments were conducted with these two systems comparing their performance with training on (1) one tempo, testing on the other, (2) one speaker, testing another, and (3) one versus several speakers. The results show that (1) a temporal description for each word can be reliably made that is invariant across speakers and speaking tempo, and (2) for these words, timing performs about as well as the commercial system but made different errors, thus suggesting that such timing information might improve recognizer performance.

Friendly interfacing to simple speech recognizers

Abstract We describe improvements to the recognition performance of a simple commercial speech recognizer. Topics include the selection of acoustically distinct words; a method of ‘training’ (storing utterances for later use as templates) which mimics the real task, and therefore reduces the difference in diction between training and task; the representation of variability in diction by storing repeated examples of each utterance separately, instead of using a simple statistical average; and the construction of an adaptive algorithm which updates its templates at appropriate moments. The results of empirical investigations with the adaptive algorithm show a very considerable improvement in performance. We argue that the development of speech recognizers has given the hardware undue attention, and that a rigorous attack on adaptive recognition, treated as a problem in control theory, would lead to a sophisticated interface to complement sophisticated hardware. The system we describe has.been successfully used in an experimental voice-operated text-editing system (Morrison and Green 1982).

Towards comprehensive but effective evaluations of the performances of commercial speech recognizers

Tests were designed for assessment of the accuracy of isolated word recognition under various controlled conditions. Desiderata guiding the design of the recommended tests included testing of all commercial recognizers by an independent agency, under identical controlled conditions that are representative of (or readily extrapolated to) expected field conditions, with all factors that affect accuracy carefully monitored, several performance measures used, and databases of sufficient size and appropriate design to assure statistical adequacy of results. Specific test conditions have been designed for speaker characteristics, training procedures, vocabularies, database recording procedures, rates of data entry, noise and channel conditions, etc. To simplify the multidimensional experimental design, and reduce the required data collections, a method of regression analysis known as “response surface methodology” has been applied. Testing is also truncated as soon as a device exhibits a substantial number of errors, so less accurate systems need less testing. Emphasis is on difficult recognition tasks that quickly reveal device limitations.