Nikolai Liubimov
About
Detail
Co-founder & CTO
United States
Résumé
Jobs
verified_user
0% verified
Chief Technology OfficerHumanSignalJun 2023 - Current (3 years 4 months)- CTOHumanSignalMay 2019 - Current (7 years 5 months)Heartex is a VC-funded startup that develops tools to label, annotate, and explore your datasets. Heartex brings AI-powered data annotation platform & open source solutions that help you to drastically reduce labeling costs, providing the ability to build new machine learning products extremely fast.
- Senior Software EngineerYandexFeb 2016 - Feb 2019 (3 years 1 month)Speech technologies & dialogue systems team. Working as backend developer for russian virtual personal assistant "Alice" and "Yandex.Station" smart speaker.
- Senior EngineerHuawei TechnologiesMar 2015 - Feb 2016 (1 year)Machine learning / data mining team
- ACTOAudio Research GroupMar 2012 - Jun 2014 (2 years 4 months)building music-related web services (retrieving, recommendation, etc.) using machine learning methods.
- scientific collaborationInstitute for Problems of Information TransmissionJun 2011 - Jun 2013 (2 years 1 month)development of pattern recognition methods in bioinformatics
- LinternshipLaboratoire dAcoustique MusicaleSep 2008 - Jan 2009 (5 months)working on audio bandwidth extension project
- software engineer, speech technology departmentSTEL Computer SystemsFeb 2007 - Mar 2015 (8 years 2 months)Development of speech, language and speaker recognition software
Education
verified_user
0% verified
- PhD Student, Applied MathematicsLomonosov Moscow State University MSUJan 2009 - Jan 2012 (3 years 1 month)
- Master's degree, Applied MathematicsLomonosov Moscow State University MSUJan 2004 - Jan 2009 (5 years 1 month)
Projects (professional or personal)
verified_user
0% verified
- TTestariumJun 2013 - Current (13 years 4 months)Testarium is a research tool to perform experiments and store results as in the repository (git or mercurial). It implements scientific template of experiments and uses numpy, colorama, flask, d3, jquery-ui and angular to provide powerful shell and beautiful presentation of your experimental data and scores.
It can be helpful for science researchers and their bosses to monitor the work progress and to optimize parameters. - WWORMTUNE: Intelligent Music PlatformMar 2012 - Current (14 years 7 months)Data mining techniques for automatical indexing and retrieval of musical tracks
- OOnline speech processing serviceJan 2010 - Current (16 years 9 months)- speech-to-text automatic transcription
- speaker identification & verification
- language recognition
- spoken dialogue system
Publications
verified_user
0% verified
- SApplication of l 1 Estimation of Gaussian Mixture Model Parameters for Language IdentificationSpeech and Computer Lecture Notes in Computer Science Vol pp SepIn this paper we explore the using of l 1 optimization for a parameter estimation of Gaussian mixture models (GMM) applied to the language identification. To train the Universal background model (UBM) at each step of Expectation maximization (EM) algorithm the problem of the GMM means estimation is stated as l 1 optimization. The approach is Iteratively reweighted least squares (IRLS). Also here is represented the corresponding solution of the Maximum a posteriori probability (MAP) adaptation. The results of the above UBM-MAP system combined with Support vector machine (SVM) are reported on the LDC and GlobalPhone datasets.
- INon-negative Matrix Factorization with Linear Constraints for Single-Channel Speech EnhancementInterspeech AugThis paper investigates a non-negative matrix factorization (NMF)-based
approach to the semi-supervised single-channel speech enhancement problem where
only non-stationary additive noise signals are given. The proposed method
relies on sinusoidal model of speech production which is integrated inside NMF
framework using linear constraints on dictionary atoms. This method is further
developed to regularize harmonic amplitudes. Simple multiplicative algorithms
are presented. The experimental evaluation was made on TIMIT corpus mixed with
various types of noise. It has been shown that the proposed method outperforms
some of the state-of-the-art noise suppression techniques in terms of
signal-to-noise - tAudio Bandwidth Extension Using Cluster Weighted Modeling of Spectral Envelopesth AES Convention OctThis paper presents a method for blind bandwidth extension of band-limited audio signals. A rough generation of the high-frequency content is performed by nonlinear distortion (waveshaping) applied to the mid-range band of the input signal. The second stage is shaping of the high-frequency spectrum envelope. It is done by a Cluster Weighted Model for MFCC coefficients, trained on full-band width audio material. An objective quality measure is introduced and the results of listening tests are presented.
- SShared latent subspace modelling within Gaussian-Binary Restricted Boltzmann Machines for NIST i-Vector Challenge 2014This paper presents a novel approach to speaker subspace modelling based on Gaussian-Binary Restricted Boltzmann Machines (GRBM). The proposed model is based on the idea of shared factors as in the Probabilistic Linear Discriminant Analysis (PLDA). GRBM hidden layer is divided into speaker and channel factors, herein the speaker factor is shared over all vectors of the speaker. Then Maximum Likelihood Parameter Estimation (MLE) for proposed model is introduced. Various new scoring techniques for speaker verification using GRBM are proposed. The results for NIST i-vector Challenge 2014 dataset are presented.