Automatic Speech Recognition on Mobile Devices and over Communication Networks

Tan, Zheng‐Hua; Lindberg, Børge

doi:10.1007/978-1-84800-143-5

Cited by 48 publications

(13 citation statements)

References 132 publications

(202 reference statements)

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Three approaches of dealing with speech recognition in embedded system with limited resources were presented in [1]. Each of these methods differs in the amount of computation being done on the embedded system.…”

Section: Isolated Speech Recognitionmentioning

confidence: 99%

“…Number of different applications for isolated speech recognition in embedded system can be proposed [1], [2]. Currently, mobile phones give people vast number of different functionalities.…”

Section: Introductionmentioning

confidence: 99%

“…On the other side, complex applications like speech recognition require high computing power and big amount of memory. In most of embedded devices it is not possible to expand memory or increase the computing power -limitations of the device are constant [1], [5], [6]. Therefore, algorithms are needed to assess the possibility of being able to realise speech recognition task on a device with given constraints.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Estimation of Hardware Requirements for Isolated Speech Recognition on an Embedded Systems

Kłobucki¹,

Maka²

2011

International Journal of Electronics and Telecommunications

View full text Add to dashboard Cite

Abstract-In recent years, speech recognition functionality is increasingly being added in embedded devices. Because of limited resources in these devices, there is a need to assess whether the defined speech recognition system is feasible within given constraints, as well as estimating how many resources the system needs. In this paper, an attempt has been taken to define a technique for estimating hardware resources usage in the speech recognition task. To determine the parameters and their dependencies in this task, the two systems were tested. The first system utilized Dynamic Time Warping pattern matching technique, the second used Hidden Markov Models. For each case, the measurement of recognition rate and time, vocabulary database size and learning time has been performed. Obtained results have been exploited to define linear and polynomial regression models, and finally, an estimation algorithm has been developed using these models. After testing proposed approach, it was observed that even low-end mobile phones have sufficient hardware resources for realisation of isolated speech recognition system.

show abstract

Section: Isolated Speech Recognitionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Estimation of Hardware Requirements for Isolated Speech Recognition on an Embedded Systems

Kłobucki¹,

Maka²

2011

International Journal of Electronics and Telecommunications

View full text Add to dashboard Cite

show abstract

“…Essa suposição permite que os termos de correlação cruzada sejam ignorados nas equações que descrevem a operação desses algoritmos, tornando a matemática mais atrativa. Por outro lado, [73][74] estabelecem que a distorção do canal h(t) torna-se aditiva no domínio log-spectral, consideradoa como um deslocamento nos parâmetros da voz limpa, sendo muitas vezes removido por meio de algum processo de filtragem linear, como por exemplo filtragem RASTA [75] ou algum outro filtro passa-banda. Com as suposições acima expostas e baseando-se nos trabalhos [76][77], será apresentada na seguinte seção a formulação matemática do modelo de distorção.…”

Section: Modelo Geral Do Ambiente Acústicounclassified

“…onde ω é o índice no domínio da DFT e H(ω) é a função de transferência do canal, que como mencionado em [73] é invariante no tempo. A partir deste ponto e para maior clareza o subíndice t será omitido, levando em consideração que as seguintes expressões aplicam-se a cada quadro.…”

Section: E Que Relaciona O Sinal De Voz X(t) Com As Distorções R(t) Eunclassified