Informatica

A large number of modern mobile devices, embedded devices and smart home devices are equipped with a voice control. Automatic recognition of the entire audio stream, however, is undesirable for the reasons of the resource consumption and privacy. Therefore, most of these devices use a voice activation system, whose task is to find the specified in advance word or phrase in the audio stream (for example, Ok, Google) and to activate the voice request processing system when it is found. The voice activation system must have the following properties: high accuracy, ability to work entirely on the device (without using remote servers), consumption of a small amount of resources (primarily CPU and RAM), noise resistance and variability of speech, as well as a small delay between the pronunciation of the key phrase and the system activation. This work is a systematic literature review on voice activation systems that satisfy the above properties. We describe the principle of various voice activation systems’ operation, the characteristic representation of sound in such systems, consider in detail the acoustic modelling and, finally, describe the approaches used to assess the models’ quality. In addition, we point to a number of open questions in this problem.

The development of Lithuanian HMM/ANN speech recognition system, which combines artificial neural networks (ANNs) and hidden Markov models (HMMs), is described in this paper. A hybrid HMM/ANN architecture was applied in the system. In this architecture, a fully connected three‐layer neural network (a multi‐layer perceptron) is trained by conventional stochastic back‐propagation algorithm to estimate the probability of 115 context‐independent phonetic categories and during recognition it is used as a state output probability estimator. The hybrid HMM/ANN speech recognition system based on Mel Frequency Cepstral Coefficients (MFCC) was developed using CSLU Toolkit. The system was tested on the VDU isolated‐word Lithuanian speech corpus and evaluated on a speaker‐independent ∼750 distinct isolated‐word recognition task. The word recognition accuracy obtained was about 86.7%.

The article presents a limited‐vocabulary speaker independent continuous Estonian speech recognition system based on hidden Markov models. The system is trained using an annotated Estonian speech database of 60 speakers, approximately 4 hours in duration. Words are modelled using clustered triphones with multiple Gaussian mixture components. The system is evaluated using a number recognition task and a simple medium‐vocabulary recognition task. The system performance is explored by employing acoustic models of increasing complexity. The number recognizer achieves an accuracy of 97%. The medium‐vocabulary system recognizes 82.9% words correctly if operating in real time. The correctness increases to 90.6% if real‐time requirement is discarded.