Home‎ > ‎

September 2013

Volume: 16, Issue: 1, September 2013

Using SNR Knowledge to Improve Noise-Robust Speech Recognition Normalization

Authors: Nathan Sherwood, Julian Cardenas Barreras, Eduardo Castillo Guerra, Julian Meng

Abstract:

We explore the benefits of using signal to noise ratio (SNR) knowledge to increase performance in noise-robust speech recognition. With this knowledge we look at the advantages behind using models built around specific SNR levels, as well as look at how we can incorporate that knowledge into feature extraction techniques. We have chosen to enhance temporal structure normalization (TSN), which normalizes the trend of noise in the features power spectral density (PSD). The performance of this technique was evaluated using the SPHINX-III speech recognition system with a modified version of the TIDIGITS speech corpus. Noise was taken from the NOIZEUS database and mixed in at varying SNR levels ranging from -5 dB to 20 dB (in increments of 5). Results show that SNR-dependent models are more effective and incorporating knowledge of noise into speech processing techniques can increase the performance of the speech recognition system with very little computational cost.

(pp. 812-815)                    Download PDF


Fuzzy-Based Recognition Model for Driving Styles

Authors: Maen Saleh, Ahmad Aljaafreh, Nashat Albdour

Abstract:

Driving styles is not only important for safety but also for fuel economy and emissions. Driving style is related to the driver behaviour as well as the driving condition. Abnormal driving behaviours can be detected by either monitoring of drivers or monitoring of vehicle-human interactions. In this paper, we propose a fuzzy system that classifies the driving styles (in-terms of vehicle-human interactions) into four classes; Below Normal (BN), Normal (N), Aggressive (A) and Very Aggressive (VA). The proposed classifier was built based on three main attributes: longitudinal acceleration, car velocity and the distance between the preceding and host car. Compared with data obtained from an expert judgment for a 25 driving styles, our proposed system shows an efficiency and reliability with a rate of 72% with the ignorance of fuzzy rules tuning process.

(pp. 816-819)                    Download PDF

Driving Styles Recognition Using Decomposed Fuzzy Logic System

Authors: Munaf Najim Al-Din, Ahmad Aljaafreh, Nashat Albdour,  Maen Saleh

Abstract:

Driver behaviour recognition or driving style classification systems have become very important for reducing vehicle accidents and providing safety. Human driver behavior recognition systems are used to estimate human behaviour by observing driver interaction with the vehicle and environment. These systems goal is to infer the driver's behaviour by mapping some parameters that represent the driver interaction with control elements available in the vehicle, such as speed, accelerator, following distance, and etc. The driving style recognition approach developed in this paper is based on the theory of fuzzy logic, which provides a methodology for reasoning about the application area that approximates human reasoning. One key aspect of the method developed in this paper is the great reduction of the rules of the rule-base inference engine by using a decomposed fuzzy system. The method is based on introducing weighting factors for the sensor inputs, thus inferring the reflexive conclusions from each input to the system rather than putting all the possible states of all the inputs to infer a single conclusion. The effectiveness and reliability of the proposed method is evaluated and compared with data obtained from an expert judgment for a 25 driving styles.

(pp. 820-824)                    Download PDF