Research Engineer, Speech Foundation Models

We are seeking a highly skilled and experienced Research Lead for Speech, Audio, and Conversational AI to join our innovative team. In this role, you will spearhead the research and development of cutting-edge technologies in speech processing, text-to-speech (TTS), audio analysis, and real-time conversational AI. You will push the boundaries of what's possible in automatic speech recognition (ASR), speaker identification, diarization, speech synthesis, voice cloning, dubbing and audio generation.

Key Responsibilities:

Bring the state of the art in Audio/Speech and Large Language Models to develop advanced Audio Language Models and Speech Language Models.

Research, architect, and deploy new generative AI methods such as autoregressive models, causal models, and diffusion models

Design and implement low-latency end-to-end models with multilingual speech/audio as both input and output.

Conduct experiments to evaluate and improve the performance of these models, focusing on accuracy, naturalness, efficiency, and real-time capabilities across multiple languages.

Stay at the forefront of advancements in speech processing, audio analysis, and large language models, integrating new techniques into our foundation models.

Collaborate with cross-functional teams to integrate these foundation models into Krutrim's AI stack and products.

Publish research findings in top-tier conferences and journals such as INTERSPEECH, ICASSP, ICLR, ICML, NeurIPS, and IEEE/ACM Transactions on Audio, Speech, and Language Processing.

Mentor and guide junior researchers and engineers, fostering a collaborative and innovative team environment.

Drive the adoption of best practices in model development, including rigorous testing, documentation, and ethical considerations in multilingual AI.

Qualifications:

Ph.D. in Computer Science, Electrical Engineering, or a related field with a focus on speech processing, audio analysis, and machine learning.

Train speech / audio models for representation (like, W2V-BERT, SONAR, AST), generation (like, Hi-Fi GAN, VQ-GAN, AudioLDM), Conformers, multilingual multitask models (like, SeamlessM4T).

Expertise with Audio Language Models like AudioPALM, Moshi and Seamless M4T

Proven track record of developing and applying novel neural network architectures such as Transformers, Mixture of Experts, Diffusion Models, and State Space Machines (MAMBA, SAMBA).

Extensive experience in developing and optimizing models for low-latency, real-time applications.

Strong background in multilingual speech recognition, voice cloning, dubbing and synthesis, with an understanding of the challenges specific to different language families.

Proficiency in deep learning frameworks (e.g., TensorFlow, PyTorch) and experience deploying large-scale speech and audio models.

Demonstrated expertise in high-performance computing with proficiency in Python, C/C++, CUDA, and kernel-level programming for AI applications.

Experience with audio signal processing techniques and their application in end-to-end neural models.