Journal
Scientific and Technical Journal of Information Technologies, Mechanics and Optics
UDK
Issue:2 (102)
Download PDF0 Kbyte
TWO-STEP ALGORITHM OF TRAINING INITIALIZATION FOR ACOUSTIC MODELS BASED ON DEEP NEURAL NETWORKS
Annotation
This paper presents a two-step initialization algorithm for training of acoustic models based on deep neural networks. The algorithm is focused on reducing the impact of the non-speech segments on the acoustic model training. The idea of the proposed algorithm is to reduce the percentage of non-speech examples in the training set. Effectiveness evaluation of the algorithm has been carried out on the example of English spontaneous telephone speech recognition (Switchboard). The application of the proposed algorithm has led to 3% relative word error rate reduction, compared with the training initialization by restricted Boltzmann machines. The results presented in the paper can be applied in the development of automatic speech recognition systems.
Keywords
Постоянный URL
Articles in current issue
- TRENDS IN THE DEVELOPMENT OF DETONATION ENGINES FOR HIGH-SPEED AEROSPACE AIRCRAFTS AND THE PROBLEM OF TRIPLE CONFIGURATIONS OF SHOCK WAVES. Part II - Research of counterpropagating shock waves and triple shock wave configurations
- CONCENTRIC LENS SYSTEMS
- EVALUATION OF CHROMATICITY COORDINATES SHIFT FOR IMAGE DISPLAYED ON LIQUID CRYSTAL PANELS WITH VARIOUS PROPERTIES ON COLOR REPRODUCTION
- DESIGNING FEATURES OF POWER OPTICAL UNITS FOR TECHNOLOGICAL EQUIPMENT
- APPLICATION OF THE LONGITUDINAL CHROMATIC ABERRATION EFFECT FOR DISTANCES MEASUREMENT ON THE BASIS OF A SINGLE PHOTO
- ANALYSIS OF CAMOUFLAGE COVER SPECTRAL CHARACTERISTICS BY IMAGING SPECTROMETER
- RESEARCH OF THERMO-OPTICAL INHOMOGENEITIES IN Yb-Er GLASS AT DIODE PUMPING
- OPTICAL PROPERTIES OF CARBAMIDE AQUEOUS SOLUTIONS
- FIRE-RESISTANCE PROPERTIES RESEARCH OF “WATER GLASS - GRAPHITE MICROPARTICLES” COMPOSITE MATERIAL
- SEMI-AUTOMATIC SPEAKER VERIFICATION SYSTEM
- COMPUTATIONALLY EFFICIENT PRIVATE INFORMATION RETRIEVAL PROTOCOL
- FORENSIC LINGUISTICS: AUTOMATIC WEB AUTHOR IDENTIFICATION
- DEVELOPMENT AND TESTING OF ERRORS CORRECTION ALGORITHM IN ELECTRONIC DESIGN AUTOMATION
- EFFICIENCY OF REDUNDANT QUERY EXECUTION IN MULTI-CHANNEL SERVICE SYSTEMS
- INVESTIGATION OF NEURAL NETWORK ALGORITHM FOR DETECTION OF NETWORK HOST ANOMALIES IN THE AUTOMATED SEARCH FOR XSS VULNERABILITIES AND SQL INJECTIONS
- EVALUATION OF SEMANTIC SIMILARITY FOR SENTENCES IN NATURAL LANGUAGE BY MATHEMATICAL STATISTICS METHODS
- INVESTIGATION OF INFLUENCE OF ENCODING FUNCTION COMPLEXITY ON DISTRIBUTION OF ERROR MASKING PROBABILITY
- IMPROVEMENT OF REFERENCE BASE IN THE FIELD OF METROLOGICAL ASSURANCE OF THREAD JOINTS
- MICROCLIMATE CARTOGRAPHY USING DATA FROM THE EARTH REMOTE SENSING AND SIMULATION OF THERMAL FIELDS
- NUMERICAL SIMULATION OF SHOCK WAVE DIFFRACTION OVER RIGHT ANGLE ON UNSTRUCTURED MESHES
- MODELING OF RAIL BAR DYNAMIC GAP AT ITS BREAK FOR DIFFERENT STIFFNESS VALUES OF RAIL BASE
- ALGEBRAIC PROPERTIES OF MATRIX COMPONENTS OF CONTROL PLANT MODELS IN PLACEMENT STATE MATRIX MODES OF SYSTEM ALGORITHMS FOR DESIGNED SYSTEM
- ANTIREFLECTION MULTILAYER COATINGS WITH THIN METAL LAYERS
- TWO-STEP ALGORITHM OF TRAINING INITIALIZATION FOR ACOUSTIC MODELS BASED ON DEEP NEURAL NETWORKS
- ON THE POSSIBILITY OF BURNING ACCELERATION IN THE COMBUSTION CHAMBERS OF ADVANCED JET ENGINES BY DEEPLY SUBCRITICAL MICROWAVE DISCHARGE