Andrej Zgank

Factors Influencing Quality of Experience

In this chapter different factors that may influence Quality of Experience (QoE) in the context of media consumption, networked services, and other electronic communication services and applications, are discussed. QoE can be subject to a range of complex and strongly interrelated factors, falling into three categories: human, system and context influence factors (IFs). With respect to Human IFs, we discuss variant and stable factors that may potentially bear an influence on QoE, either for low-level (bottom-up) or higher-level (top-down) cognitive processing. System IFs are classified into four distinct categories, namely content-, media-, network- and device-related IFs. Finally, the broad category of possible Context IFs is decomposed into factors linked to the physical, temporal, social, economic, task and technical information context. The overview given here illustrates the complexity of QoE and the broad range of aspects that potentially have a major influence on it.

Clustering of triphones using phoneme similarity estimation for the definition of a multilingual set of triphones

This paper addresses the problem of multilingual acoustic modelling for the design of multilingual speech recognisers. An agglomerative clustering algorithm for the definition of multilingual set of triphones is proposed. This clustering algorithm is based on the definition of an indirect distance measure for triphones defined as a weighted sum of the explicit estimates of the context similarity on a monophone level. The monophone similarity estimation method is based on the algorithm of Houtgast. The new clustering algorithm was tested in a multilingual speech recognition experiment for three languages. The algorithm was applied on monolingual triphone sets of language specific recognisers for all languages. In order to evaluate the clustering algorithm, the performance of the multilingual set of triphones was compared to the performance of the reference system composed of all three language specific recognisers operating in parallel, and to the performance of the multilingual set of triphones produced by the tree-based clustering algorithm. All experiments were based on the 1000 FDB SpeechDat(II) databases (Slovenian, Spanish and German). Experiments have shown that the use of the clustering algorithm results in a significant reduction of the number of triphones with minor degradation of recognition rate.

Can context monitoring improve QoE? A case study of video flash crowds in the internet of services

Over the last decade or so, significant research has focused on defining Quality of Experience (QoE) of Multimedia Systems and identifying the key factors that collectively determine it. Some consensus thus exists as to the role of System Factors, Human Factors and Context Factors. In this paper, the notion of context is broadened to include information gleaned from simultaneous out-of-band channels, such as social network trend analytics, that can be used if interpreted in a timely manner, to help further optimise QoE. A case study involving simulation of HTTP adaptive streaming (HAS) and load balancing in a content distribution network (CDN) in a flash crowd scenario is presented with encouraging results.

Large Vocabulary Continuous Speech Recognizer for Slovenian Language

The paper describes the development of a large vocabulary continuous speech recogniser for Slovenian language with SNABI database. The problems with inflectional languages when speech recognition is performed are presented. The system is based on hidden Markov models. For acoustic modeling biphones were used whereas for language modeling bigrams and trigrams were used. To improve the recognition result and to enable fast operation of the recogniser, speaker adaptation is also used. The optimal system with the adapted acoustic model and bigram language model achieved word accuracy of 91.30% at near 10x real time. The unadapted system with the trigram language model achieved the word accuracy of 89.56%, but it was also slower than the optimal system. Its run time was 15.3x real time.

Comparison of Acoustic Adaptation Methods in Multilingual Speech Recognition Environment

This paper presents the comparison of different acoustic adaptation methods in a multilingual speech recognition environment. Baseline multilingual acoustic models were generated using the tree based clustering with common phonetic broad classes. After the expert based port to a new language was performed, the influence of several adaptation methods on speech recognition performance was investigated. The target language adaptation subset contained 2% of complete speech database. The best adapted ported system had significant improvement in the speech recognition performance and its results were close to the results of pure reference monolingual system. The relationship between languages used in the mapping configuration remained unchanged after the adaptation. ...

Context Monitoring for Improved System Performance and QoE

Whereas some application domains show a certain consensus on the role of system factors, human factors, and context factors, QoE management of multimedia systems and services is still faced with the challenge of identifying the key QoE influence factors. In this chapter, we focus on the potential of enhancing QoE management mechanisms by exploiting valuable context information.

Lag Compensation for First-Person Shooter Games in Cloud Gaming

Cloud gaming is an emerging technology that combines cloud computing with computer games. Compared to traditional gaming, its core advantages include ease of development/deployment for developers, and lower technology costs for users given the potential to play on thin client devices. In this chapter, we firstly describe the approach, and then focus on the impact of latency, known as lag, on Quality of Experience, for so-called First Person Shooter games. We outline our approach to lag compensation whereby we equalize within reason the up and downlink delays in real-time for all players. We describe the testbed in detail, the open source Gaming Anywhere platform, the use of NTP to synchronise time, the network emulator and the role of the centralized log server. We then present results that firstly validate the mechanism and also use small scale and preliminary subjective tests to assess and prove its performance. We conclude the chapter by outlining ongoing and future work.

Analyzing the influence of narrow-band channel on Slovenian broadcast news speech recognition

The aim of this paper was to analyze the influence of narrow-band input channel on a broadcast news speech recognition system. Different acoustic conditions can be found within a typical broadcast news domain, where narrow-band channel presents one of those with possible high impact on accuracy. A method for acoustic channel detection, based on HMM models is proposed, in order to distinguish between the input channels. The advantage of this method is its low system complexity. The Slovenian BNSI Broadcast News and SNABI speech databases were used for the experimental setup. The Slovenian UMB Broadcast News automatic speech recognizer was applied as a test-bed, modified appropriately for the task. The evaluation of HMM models for channel detection showed accuracy higher than 90% for both channel types. The channel influence analysis confirmed that narrow-band input channel significantly degrades the speech recognition accuracy, decreasing it by more than 13% absolute.

Uniform Speech Recognition Platform for Evaluation of New Algorithms

This paper presents the development of a speech recognition platform. Its main area of use would be the evaluation of different new and improved algorithms for speech recognition (noise reduction, feature extraction, language model generation, training of acoustic models, ...). To enable wide usage of the platform, different test configurations were added - from alphabet spelling to large vocabulary continuous speech recognition. At the moment, this speech recognition platform is implemented and evaluated using a studio (SNABI) and a fixed telephone (SpeechDat(II)) speech database.