Research & Development — Label Studio, HumanSignal & speech recognition
Label Studio & HumanSignal
Open-source data labeling and an AI platform [2019–today]I'm a co-founder of HumanSignal and of Label Studio, the open-source tool teams use to annotate, label and explore datasets and get them ready for machine learning.
Label Studio has 27k+ stars on GitHub and 1M+ users, and is trusted by teams at Cloudflare, NVIDIA, Meta, IBM and Intel. The enterprise platform is SOC 2 Type II and HIPAA compliant. We raised a $25M Series A led by Redpoint Ventures to build it out.
Speech recognition solutions
Language and speaker identification, diarization [2011]I run a lot of R&D projects at ASM Solutions. TensorFlow, PyTorch, GMMs, i-vectors and filterbanks are part of my daily practice. I attend international conferences in the field and publish there. All of the research runs in Python and C++, and we also build a speech-processing SDK and high-performance computing solutions with CUDA.

TfMicro: Lightweight framework for TensorFlow
Build your TF models comfortably [2017–2019]
Focus on the model — TfMicro handles the rest: a multiprocessing data pipeline, callbacks, keyboard controls, and saving / loading / preloading of models, and much more.
TfMicro on GitHub

Testarium: Research tool, open source
Run experiments and store the results in a repository [2014–2019]It provides a scientific template for experiments and uses NumPy, Colorama, Flask, D3, jQuery UI and Angular to give you a powerful shell and a clean presentation of your experimental data and scores.
It helps researchers and their managers track progress and tune parameters. I used it in my own work on neural networks and speech recognition. See testarium.makseq.com.

ECG Demodulator and iOS/Android Mobile App
FM Demodulator for the electrocardiogram signals (ECG) [2013-2014]iOS/Android mobile app for ECG recording and management [2014-2015]
I designed the FM demodulator software for BIOSS ECG devices. It's used by KardioDom in their healthcare app, HelterBook.
Later, my team and I designed and built an iOS / Android app for KardioDom on top of this demodulator.
Get it on Google Play

Publications
Here is the list of my scientific publications (part of them are in russian only):- Lightweight embeddings for speaker verification. Maxim Tkachenko, Alexander Yamshinin, Mikhail Kotov. Speecom, 2018 link
- Speech Enhancement for Speaker Recognition Using Deep Recurrent Neural Networks. Maxim Tkachenko, Alexander Yamshinin, Nikolay Luibimov, Mikhail Kotov. Speecom, 2017 link
- Language Identification using Time Delay Neural Network D-Vector on Short Utterances. Maxim Tkachenko, Alexander Yamshinin, Nikolay Luibimov, Mikhail Kotov. Speecom, 2016 link
- Danila Doroshin, Alexander Yamshinin, Nikolay Lubimov, Maxim Tkachenko. Shared latent subspace modelling within Gaussian-Binary Restricted Boltzmann Machines for NIST i-Vector Challenge 2014. Interspeech, 2015 link
- Danila Doroshin, Maxim Tkachenko, Nikolay Lubimov, Mikhail Kotov. Application of l_1 Estimation of Gaussian Mixture Model Parameters for Language Identification. Speech and Computer Lecture Notes in Computer Science Volume 8113, 2013, pp 41-45 link
- Maxim S. Tkachenko. Modeling Environment for Particles Configuration Control in Magnetic and Electric Fields. Intel International Science and Engineering Fair. Phoenix, 2005
- М.С. Ткаченко. Среда для моделирования вычислительных задач. // Журнал Мир ПК – Диск, 2006, №4 (40), (2006a04_cd05.pdf), c.1-5.
- М.С. Ткаченко, А.М. Умнов, А.В. Гиглавый. Инструментальная среда моделирования и визуализации исследования динамики электронов в магнитных и электрических полях. // В сб.: Материалы 28 Всероссийского семинара “Озон и другие экологически чистые окислители. Наука и технологии”. Москва, МГУ, 2 июня 2006, с. 124.
- М.С. Ткаченко, А.В. Гиглавый, А.М. Умнов. Generation ISEF или рекомендации по подготовке и участию в Intel International Science and Engineering Fair на примере работы “Инструментальная среда моделирования и визуализации движения заряженных частиц в магнитных и электрических полях”. // Москва, изд-во МАКС Пресс, 2007, - 40 с. (Приложение: CD диск с примерами.)
- А.М. Умнов, М.С. Ткаченко. Визуализация движения заряженных частиц в электромагнитных и магнитостатических полях в интерактивных обучающих программах. // В сб. трудов Международной научно-практической конференции “Образовательные, научные и инженерные приложения в среде LabVIEW и технологии National Instruments”. Москва, 14-15 ноября 2003 г. –М.: изд-во РУДН, 2003, с.107-109.
- М.С. Ткаченко, А.М. Умнов. Инструментальная среда визуализации и исследования движения заряженных частиц в магнитных и электрических полях. // В сб. аннотаций докладов по озонной тематике, представленных на семинарах в МЭИ и МГУ им. М.В. Ломоносова (1994-2008 гг.). Москва, изд-во МАКС Пресс, 2008, с. 87-88.
- М.С. Ткаченко, А.М. Умнов. Инструментальная среда визуализации и исследования движения заряженных частиц в магнитных и электрических полях. // В сб. материалов 30-й Всероссийский семинар "Озон и другие экологически чистые окислители. Наука и технологии". Москва, изд-во МАКС Пресс, 2008, с. 301-302.
- М.С. Ткаченко. Инструментальная среда моделирования и визуализации движения заряженных частиц в магнитных и электрических полях. // В сб. тезисов "Ломоносов-2008" секция "Физика". Москва, изд-во отдела оперативной печати физического факультета, 2008, с. 99-100.
- М.С. Ткаченко, А.М. Умнов. Научно-практический журнал "Открытое образование". 6(77)'2009. c. 37-43.
- М.С. Ткаченко, А.С. Лукин. Многомасштабный метод спектрального вычитания для подавления шумов в аудиосигналах. В сб. тезисов "DSPA-2010". Москва. 2010. с. 124-125
- М.С. Ткаченко. Многомасштабный метод подавления шумов в аудиосигналах. Сборник тезисов лучших дипломных работ 2010. Московский государственный университет имени М.В. Ломоносова, факультет вычислительной математики и кибернетики. Москва, 2010, с. 20.
- М.С. Ткаченко. Научно-познавательная статья: «Как слушать музыку», 2010.
Read moreShow less
Loopseque/MovBeats. Audio/Video sequencers for iOS
Loopseque - audio sequencer for iOS [2010-2012]MovBeats - video sequencer for iOS [2011-2012]
Maths and programming of audio processing, sound engine architecture, concept & ideas generation, audio routing scheme, some of UI programming, asm maths optimizations for ARMv6 processors - all of them were mine.
MovBeats is using my audio engine from Loopseque with little changes in audio routing scheme.
MovBeats official web site
Adaptive noise reduction system based on multi-resolution FFT
Online audio denoising service [2007-2009]My master's thesis later grew into an online service. I've since discontinued it and taken the service offline.

Tool for plasma modeling and visualization
Modeling environment for particle-configuration controlin magnetic and electric fields [2003-2006]
My first R&D project won a third award at Intel ISEF.
You can download it here (Win x86)
Read moreShow less
A toolkit like this makes the processes under study easier to understand and helps researchers obtain meaningful results in physics and its applications.
The project is based on simulation of a pre-defined space-time configuration of charged particles in order to study their behavior in electric and magnetic fields. To provide universality, flexibility and processing speed of the proposed modeling toolkit a wide range of software tools has been applied. Flexible and user-friendly program interface has been based on Borland Delphi. All calculations are carried out by means of a program module written in Compaq Fortran. Visualizing modules initially written in Delphi have been revised in Microsoft Visual Studio C++ with the use of OpenGL library. As a result the rendering speed has substantially increased.