Senior Automatic Speech Recognition (ASR) Engineer

emagine; Bengaluru; Presencial; INR 3 000 000 - 5 000 000; Tempo integral

Resumo da oferta

emagine is seeking a Senior Automatic Speech Recognition (ASR) Engineer to join its ASR Research & Development team. You will work on challenging research problems and develop state-of-the-art speech recognition technologies across multiple languages, combining research, experimentation, and hands-on model development to improve ASR accuracy.

Experience with Conformer-based ASR models is highly desirable; you will train, optimise, and evaluate end-to-end systems, curate data, and collaborate

Qualificações

• 5+ years in speech technology, ASR, or related R&D.

Responsabilidades

• Research, develop, train, and optimise end-to-end ASR models to improve accuracy and robustness.
• Conduct experiments to evaluate ASR model quality and analyse the impact of hyperparameters, training strategies, and model configurations.
• Develop and improve model training recipes, fine‑tuning strategies, and language modelling techniques.
• Work with Conformer-based architectures and other state-of-the‑art speech recognition models, including model training and optimisation where applicable.
• Identify, curate, clean, optimise, and cluster speech and text training datasets to improve model performance.
• Analyse large‑scale product usage data and speech recognition errors to identify performance gaps and opportunities for improvement.
• Apply model optimisation techniques, including pruning, efficient representations, model compression, and language model interpolation.
• Improve ASR performance across multiple languages, accents, speaking styles, and real‑world acoustic conditions.
• Stay up to date with advancements in speech technology, deep learning, computational linguistics, and natural language processing.
• Design and execute research experiments, interpret results, and present findings and recommendations to the research team.
• Collaborate with cross‑functional engineering, research, and product teams to translate business requirements into effective technical solutions.
• Maintain high standards of code quality, reproducibility, experimentation, and documentation.

Conhecimentos

ASR research

Conformer models

End-to-end ASR

Python

Shell scripting

Linux

Model evaluation

Hyperparameter tuning

ML fundamentals

English communication

Formação académica

MSc or PhD in Computer Science, Computational Linguistics, Applied Mathematics, Electrical Engineering, or related

Ferramentas

ESPnet

NVIDIA NeMo

SpeechBrain

Kaldi

Descrição da oferta de emprego

Summary

We are looking for a Senior Automatic Speech Recognition (ASR) Engineer to join our ASR Research & Development team. In this role, you will work on challenging research problems and develop state-of-the‑art speech recognition technologies that improve real‑world user experiences across multiple languages. You will combine research, experimentation, and hands‑on model development to improve ASR accuracy, optimise training pipelines, and enhance language modelling capabilities. You will evaluate emerging algorithms, analyse large‑scale speech and product usage data, and contribute to the continuous improvement of model architectures and training recipes. Experience training and optimising Conformer‑based ASR models will be highly desirable.

Key Responsibilities

• Research, develop, train, and optimise end-to-end ASR models to improve speech recognition accuracy and robustness.
• Conduct experiments to evaluate ASR model quality and analyse the impact of hyperparameters, training strategies, and model configurations.
• Develop and improve model training recipes, fine‑tuning strategies, and language modelling techniques.
• Work with Conformer-based architectures and other state-of-the‑art speech recognition models, including model training and optimisation where applicable.
• Identify, curate, clean, optimise, and cluster speech and text training datasets to improve model performance.
• Analyse large‑scale product usage data and speech recognition errors to identify performance gaps and opportunities for improvement.
• Apply model optimisation techniques, including pruning, efficient representations, model compression, and language model interpolation.
• Improve ASR performance across multiple languages, accents, speaking styles, and real‑world acoustic conditions.
• Stay up to date with advancements in speech technology, deep learning, computational linguistics, and natural language processing.
• Design and execute research experiments, interpret results, and present findings and recommendations to the research team.
• Collaborate with cross‑functional engineering, research, and product teams to translate business requirements into effective technical solutions.
• Maintain high standards of code quality, reproducibility, experimentation, and documentation.

Required Skills And Qualifications

• MSc or PhD in Computer Science, Computational Linguistics, Applied Mathematics, Electrical Engineering, or a related discipline.
• 5+ years of relevant experience in speech technology, ASR, machine learning, or a closely related research and development field.
• Strong background in speech recognition, computational linguistics, machine translation, and/or natural language processing.
• Extensive hands‑on experience with end‑to‑end ASR systems and language modelling.
• Strong understanding of deep learning architectures, machine learning principles, statistical methods, and model evaluation techniques.
• Practical experience with model training, fine‑tuning, hyperparameter optimisation, and performance evaluation.
• Proficiency in Python and shell scripting, with strong software engineering fundamentals.
• Experience working in Linux environments; familiarity with Windows-based development environments.
• Ability to analyse experimental results, investigate model errors, and identify effective solutions to complex technical problems.
• Strong analytical, problem‑solving, and critical‑thinking skills.
• Excellent written and verbal communication skills in English.

Preferred Skills

• Experience training, fine‑tuning, and optimising Conformer‑based ASR models.
• Experience with Transformer-based architectures, attention mechanisms, sequence modelling, and modern end‑to‑end speech recognition approaches.
• Knowledge of acoustic modelling, pronunciation modelling, decoding strategies, and speech recognition evaluation metrics such as Word Error Rate (WER) and Character Error Rate (CER).
• Experience with large‑scale speech data processing, dataset selection, data augmentation, and multilingual speech modelling.
• Experience with generative AI, Large Language Models (LLMs), prompt engineering, and fine‑tuning techniques.
• Expertise in one or more areas such as machine learning, statistical modelling, information retrieval, deep learning, or data mining.
• Experience with model compression, pruning, quantisation, efficient inference, or deployment optimisation.
• Familiarity with ASR frameworks and tools such as ESPnet, NVIDIA NeMo, SpeechBrain, Kaldi, or equivalent technologies.
• Fluency in a second language other than English.