A comparative review of dynamic neural networks and hidden Markov model methods for mobile on-device speech recognition

Mustafa, Mohammed Kyari; Allen, Tony; Appiah, Kofi

doi:10.1007/s00521-017-3028-2

Cited by 70 publications

(38 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Although, as the network size increases for DMLP, the computation time will increase. But this increase is significantly lower than an HMM system [14].…”

Section: Literature Reviewmentioning

confidence: 79%

ASR System for Isolated Words using ANN with Back Propagation and Fuzzy based DWT

Mendiratta¹,

Turk²,

Bansal³

2019

IJEAT

View full text Add to dashboard Cite

Speech is the primary means through which human beings interact. Speech has become a way for Man Machine Interaction (MMI). The Speech Recognition (SR) systems have been widely used in smart phones to initiate searches or to type certain text messages, and in control devices to perform switch on or off functions etc. This system comprises three blocks: Pre-processing, Feature Extraction and Classification. The input speech signal is pre-processed to remove the noise and to convert it into a digital form for feature extraction. The feature extraction is a significant process during SR systems design because the features extracted form the basis for accurate recognition of the speech. Only a few features of this signal may be selected for classification purposes. For final recognition of the spoken word or the input signal, various optimization algorithms as classifiers are used. This paper presents an extensive literature review on SR Systems. The authors have attempted to do a brief survey to identify the progress in this field. The survey provides the reader with well-known methods used by previous researchers. It also compares the performance metrics for two ASR techniques developed by the authors. The first technique uses Artificial Neural Network with Back Propagation while the second uses Fuzzy based Discrete Wavelet Transform. It was found that the fuzzy based DWT system provided better results in terms of the performance metrics like accuracy, sensitivity, specificity and word error rate. The paper concludes by providing the reader with a direction of future scope in this research area.

show abstract

“…Although, as the network size increases for DMLP, the computation time will increase. But this increase is significantly lower than an HMM system [14].…”

Section: Literature Reviewmentioning

confidence: 79%

ASR System for Isolated Words using ANN with Back Propagation and Fuzzy based DWT

Mendiratta¹,

Turk²,

Bansal³

2019

IJEAT

View full text Add to dashboard Cite

show abstract

“…There are metrics meant to evaluate the impact of speech in a room, such as the speech intelligibility index, but such measurements are better at characterizing the acoustic properties of a space rather than quantifying the impact of actual human-generated sounds in time. Machine learning algorithms are a promising solution to this dilemma, as audio can be analyzed in real time for context and emotional content, as well as decipher speech-generated sounds from other background sounds [88,89].…”

Section: Auditory Conditionsmentioning

confidence: 99%

The Spatial and Temporal Variability of the Indoor Environmental Quality during Three Simulated Office Studies at a Living Lab

et al. 2019

View full text Add to dashboard Cite

The living lab approach to building science research provides the ability to accurately monitor occupants and their environment and use the resulting data to evaluate the impact that various components of the built environment have on human comfort, health, and well-being. A hypothesized benefit of the living lab approach is the ability to simulate the real indoor environment in an experimentally controlled setting over relatively long periods of time, overcoming a significant hurdle encountered in many chamber-type experimental designs that rarely reflect typical indoor environments. Here, we present indoor environmental quality measurements from a network of sensors as well as building system design and operational data demonstrating the ability of a living lab to realistically simulate a wide range of environmental conditions in an office setting by varying air temperature, lighting, façade control, and sound masking in a series of three human subject experiments. The temporal variability of thermal and lighting conditions was assessed on an hourly basis and demonstrated the significant impact of façade design and control on desk-level measurements of both factors. Additional factors, such as desk layout and building system design (e.g., luminaires, speaker system), also contributed significantly to spatial variability in air temperature, lighting, and sound masking exposures, and this variability was reduced in latter experiments by optimizing desk layout and building system design. While ecologically valid experimental conditions are possible with a living lab, a compromise between realism and consistency in participant experience must often be found by, for example, using an atypical desk layout to reduce spatial variability in natural light exposure. Based on the experiences from these three studies, experimental design and environmental monitoring considerations for future office-based living lab experiments are explored.

show abstract

“…More complex software for Orthoptera song identification is based on Artificial Neural Networks (ANN) and Hidden Markov Models (HMM) which are widely used in automatic human speech recognition (Mustafa et al 2017). Because ANNs have to be trained by a set of training recordings, and later be tested on another validation set, this approach is only possible for identification of species with at least ten recordings of distinct specimens.…”

Section: The Way Forward: Algorithms For Acoustic Profilingmentioning

confidence: 99%

Acoustic profiling of Orthoptera: present state and future needs

Riede¹

2018

JOR

View full text Add to dashboard Cite

Bioacoustic monitoring and classification of animal communication signals has developed into a powerful tool for measuring and monitoring species diversity within complex communities and habitats. The high number of stridulating species among Orthoptera allows their detection and classification in a non-invasive and economic way, particularly in habitats where visual observations are difficult or even impossible, such as tropical rainforests. Major sound archives were queried for Orthoptera songs, with special emphasis on usability as reference training libraries for computer algorithms. Orthoptera songs are highly stereotyped, reliable taxonomic features. However, exploitation of songs for acoustic profiling is limited by the small number of reference recordings: existing song libraries represent only about 1000 species, mainly from Europe and North America, covering less than 10% of extant stridulating Orthoptera species. Available databases are fragmented and lack tools for song annotation and efficient feature-based searching. Results from recent bioacoustic surveys illustrate the potential of the method, but also the challenges and bottlenecks impeding further progress. A major problem is time-consuming data analysis of recordings. Computer-aided identification software exists for classification and identification of cricket and grasshopper songs, but these tools are still far from practical for field application.A framework for acoustic profiling of Orthoptera should consist of the following components: (1) Protocols for standardized acoustic sampling, at species and community levels, using acoustic data loggers for autonomous long-term recordings; (2) Open access to and efficient management of song data and voucher specimens, involving the Orthoptera Species File (OSF) and Global Biodiversity Information Facility (GBIF); (3) An infrastructure for automatized analysis and song classification; and (4) Complementation and improvement of Orthoptera sound libraries using OSF as the taxonomic backbone and repository for representative song recordings. Taxonomists should be encouraged, or even obliged, to deposit original recordings, particularly if they form part of species descriptions or revisions. Citation: Riede K (2018) Acoustic profiling of Orthoptera: present state and future needs. Journal of Orthoptera Research 27(2): 203-215. https://doi.

show abstract

A comparative review of dynamic neural networks and hidden Markov model methods for mobile on-device speech recognition

Cited by 70 publications

References 12 publications

ASR System for Isolated Words using ANN with Back Propagation and Fuzzy based DWT

ASR System for Isolated Words using ANN with Back Propagation and Fuzzy based DWT

The Spatial and Temporal Variability of the Indoor Environmental Quality during Three Simulated Office Studies at a Living Lab

Acoustic profiling of Orthoptera: present state and future needs

Contact Info

Product

Resources

About