Learn More
Hidden Markov model speech recognition systems typically use Gaussian mixture models to estimate the distributions of decor-related acoustic feature vectors that correspond to individual sub-word units. By contrast, hybrid connectionist-HMM systems use discriminatively-trained neural networks to estimate the probability distribution among subword units(More)
Hidden Markov model speech recognition systems typically use Gaussian mixture models to estimate the distributions of decor-related acoustic feature vectors that correspond to individual sub-word units. By contrast, hybrid connectionist-HMM systems use discriminatively-trained neural networks to estimate the probability distribution among subword units(More)
We evaluate the performance of several feature sets on the AURORA task as defined by ETSI. We show that after a non-linear transformation, a number of features can be effectively used in a HMM-based recognition system. The non-linear transformation is computed using a neural network which is discriminatively trained on the phonetically labeled (forcibly(More)
OBJECTIVE While corner store-based nutrition interventions have emerged as a potential strategy to increase healthy food availability in low-income communities, few evaluation studies exist. We present the results of a trial in Baltimore City to increase the availability and sales of healthier food options in local stores. DESIGN Quasi-experimental study.(More)
The mutual information concept is used to study the distribution of speech information in frequency and in time. The main focus is on the information that is relevant for phonetic classiication. A large database of hand-labeled uent s p e e c h is used to (a) compute the mutual information (MI) between a phonetic classiication variable and one spectral(More)
Rather long temporal trajectory of critical band logarithmic power spectrum energy at a given frequency is used as an input feature vector in a MLP-based phoneme classiier, trained on a task-independent hand-labeled development data. Class-speciic log likelihood vectors from the individual sub-classiiers form input to a merging MLP classiier trained on the(More)
The aims of the present study were to (1) characterise the diets of adult Inuit; (2) highlight foods for a nutritional and lifestyle intervention programme; (3) develop a quantitative FFQ (QFFQ) to evaluate the programme and monitor changes in dietary intake in this population over time. A dietary survey using single 24-h dietary recalls was conducted among(More)
A common approach to measuring the impact of noise and the effectiveness of noise mitigation (NM) algorithms for Automatic Speech Recognition (ASR) systems is to compare the word error rates (WERs). However, the WER measure does not give much insight into how an NM algorithm affects phoneme-level acoustic characteristics. Such insight can help in tuning the(More)