Assel Yermekova

ML Researcher (Middle → Senior)

Huawei Noah’s Ark Laboratory

Huawei's research laboratory focused on AI and related technologies.

Led speech and audio ML research, building evaluation and data pipelines and developing methods for captioning, TTS, tokenization, and voice conversion.

• Led and initiated research on automatic audio captioning (AAC) evaluation; developed internal benchmark to analyze 15 AAC metrics reliability. Built AAC dataset (7,300+ hours) and improved captioning quality (CIDEr-D: +13.09 → 70.25) on internal Omni LLM. • Designed Speech LLM automatic data collection pipeline of 40k+ hours of SFT data. Designed and implemented evaluation pipeline for SpeechLLMs on 11 benchmarks with web dashboards for tracking model performance.

Research Assistant

Nazarbaev University

A research university in Kazakhstan.

Led a multimodal speaker verification project and developed an encoder that improved visual and audio verification against unimodal baselines.

• Led multimodal speaker verification project (RGB, thermal, audio). • Designed universal encoder trained on prototypical loss and improved verification performance for visual modality on 77.6 % and for audio on 18.58 % compared to unimodal baselines.

Research Engineer

Skoltech

Skolkovo Institute of Science and Technology, a graduate research university in Russia.

Built Gaussian Process regression models for air pollution source detection, accounting for spatial dependencies and uncertainty.

• Designed and implemented Gaussian Process regression models for air pollution source detection, capturing spatial dependencies and uncertainty in environmental data.

Lead of Speech Processing Course

Deep Learning School

Led a speech processing course.

Mentor of Kazakhstan team

IOAI International Olympiad

Mentored the Kazakhstan team at the IOAI International Olympiad.

MSc, Information Science and Technology

Skolkovo Institute of Science and Technology

BSc, Physics

Eurasian National University