![Scientific and Technical Journal of Information Technologies, Mechanics and Optics](/images/mag-ntv.png)
KNOWLEDGE TRANSFER FOR RUSSIAN CONVERSATIONAL TELEPHONE AUTOMATIC SPEECH RECOGNITION
![Scientific and Technical Journal of Information Technologies, Mechanics and Optics](/images/mag-ntv.png)
Annotation
This paper describes the method of knowledge transfer between the ensemble of neural network acoustic models and student-network. This method is used to reduce computational costs and improve the quality of the speech recognition system. The experiments consider two variants of generation of class labels from the ensemble of models: interpolation with alignment, and the posteriori probabilities. Also, the quality of models was studied in relation with the smoothing coefficient. This coefficient was built into the output log-linear classifier of the neural network (softmax layer) and was used both in the ensemble and in the student-network. Additionally, the initial and final learning rates were analyzed. We were successful in relationship establishing between the usage of the smoothing coefficient for generation of the posteriori probabilities and the parameters of the learning rate. Finally, the application of the knowledge transfer for the automatic recognition of Russian conversational telephone speech gave the possibility to reduce the WER (Word Error Rate) by 2.49%, in comparison with the model trained on alignment from the ensemble of neural networks.
Keywords
Постоянный URL
Articles in current issue
- ON MODERN APPROACH TO AIRPLANE-TYPE UNMANNED AERIAL VEHICLES DESIGN WITH SHORT TAKEOFF AND LANDING PART III. NUMERICAL MODELING OF AIRCRAFT VORTEX AERODYNAMICS BY DISCRETE VORTEX METHOD
- LOCALIZED LASER VAPORIZATION OF FILMS WITH COMPLEX TOPOLOGIES FOR SURFACE ACOUSTIC WAVE MICROGYROSCOPE SENSOR
- OPTICAL FIELD AMPLITUDE DISTRIBUTION ON THE PATTERN PLATE OF OPTOELECTRONIC SYSTEM FOR MEASURING OF DITHER SYSTEM PARAMETERS IN RING LASER GYRO
- CHOOSING PARAMETERS OF SPATIAL POSITION CONTROL OPTICAL-ELECTRONIC SYSTEMS WITH ACTIVE REFERENCE MARKS
- DEFORMATION CONTROL METHOD OF COMPOSITE STRUCTURAL ELEMENTS BY FIBER-OPTIC ACOUSTIC EMISSION SENSOR
- SEMICONDUCTOR FREQUENCY STANDARD BASED ON P(16) SPECTRAL LINE OF ACETYLENE ISOTOPE WITH TEMPERATURE STABILIZATION BY PHASE MODULATION
- ELECTRIC GENERATOR CONTROL UNDER HIGH-FREQUENCY MEASUREMENT NOISES
- AUDIO-VISUAL SPEECH PROCESSING AND ANALYSIS BASED ON SUBSPACE PROJECTIONS
- EFFICIENCY IMPROVEMENT OF CODING METHOD BY INTRAFRAME PREDICTION IN H.265 / HEVC STANDARD
- DATABASE SEMANTIC MODEL APPLICATION IN NATURAL LANGUAGE USER INTERFACE DEVELOPMENT PROCESS
- INDUSTRY 4.0 DIGITAL PRODUCTION ORGANIZATION BASED ON CYBER AND PHYSICAL SYSTEMS AND ONTOLOGIES
- INFORMATIVE FEATURE SELECTION IN SOFTWARE IDENTIFICATION TASK
- DESIGN, DEVELOPMENT AND MAINTENANCE METHODOLOGY OF DOMAIN SEMANTIC PORTALS OF SCIENTIFIC AND TECHNICAL INFORMATION
- APPLICATION OF MAСHINE LEARNING METHODS FOR DETECTING OF JPEG IMAGE INTEGRITY VIOLATIONS
- OBJECT-PROCESS DATA MODEL FOR SERVICE-ORIENTED ARCHITECTURE OF INTEGRATED INFORMATION SYSTEMS
- ANALYSIS OF USERS’ PROTECTION FROM SOCIO-ENGINEERING ATTACKS: SOCIAL GRAPH CREATION BASED ON INFORMATION FROM SOCIAL NETWORK WEBSITES
- OPTIMAL MATHEMATICAL MODEL FOR DESCRIPTION OF PHYSICAL PHENOMENA AND TECHNOLOGICAL PROCESSES
- SYNTHESIS METHOD OF DIGITAL-TO-ANALOG CONVERTER SCHEMATIC MODELS FOR INTEGRATED CIRCUITS
- INTEGRATED CIRCUITS TIMING ANALYSIS WITH ACCOUNT OF PAD MODELS AND BOND WIRES
- ACCURACY INCREASE FOR AUTOMATIC VISUAL RUSSIAN SPEECH RECOGNITION: VISEME CLASSES OPTIMIZATION
- FEATURE COMBINATION FOR THE TASK OF NEURAL NETWORK ACOUSTIC MODEL LEARNING