Applied Researcher, Audio

nyra health

3d•Hybrid

About The Position

As an Applied Researcher in Audio, you will turn promising research into models that work outside the lab. You will contribute across model architecture, data, training, evaluation, and inference. Depending on the problem, your work could involve speech understanding, generation, representation learning, alignment, multilingual modeling, or multimodal systems. This role is deliberately broad. We are looking for someone who can move between scientific exploration and practical implementation, then carry a successful experiment through to an open release or production system. Audio contains much more than the words in a transcript. Timing, prosody, speaker identity, pronunciation, repairs, vocal events, and acoustic context all carry information. Most speech systems simplify these details away. That makes them easier to train, but less useful in real communication and especially in neurological care. nyra labs works on models that preserve and understand more of the original signal. We need an applied researcher who can connect new research ideas with difficult real-world data, rigorous evaluation, and systems that people can actually use.

Requirements

A strong background in speech, audio understanding, audio generation, speech-to-speech systems, or representation learning.
You balance scientific novelty with usefulness and measurable impact.
Hands-on experience with PyTorch, modern model architectures, and large-scale training.
You are comfortable working across architecture, data, evaluation, and infrastructure.
You design informative experiments, choose meaningful metrics, and interpret results carefully.
You write clean Python and can move beyond notebooks into maintainable systems.
MSc, PhD, or equivalent practical experience in machine learning, speech processing, audio, or a related field.
You use modern research and coding tools to accelerate exploration, implementation, and analysis.

Nice To Haves

You are willing to follow the problem across disciplinary boundaries.
You know when a simple baseline is more informative than a complicated model.
You want research to reach users, not stop at a benchmark.
You enjoy working with researchers, engineers, therapists, and product teams.
You can identify the next useful experiment and make it happen.

Responsibilities

Research and develop models for speech understanding, generation, alignment, representation learning, and related areas.
Explore architectures that can reason across audio, text, timing, and other relevant signals.
Curate training mixtures, improve annotation methods, and develop synthetic or model-assisted data pipelines.
Establish benchmarks that measure the details conventional audio metrics miss.
Move quickly from papers and hypotheses to working experiments and clear conclusions.
Train and optimize models efficiently across modern GPU infrastructure.
Work with engineering to turn successful prototypes into reliable open models and nyra health capabilities.
Contribute to papers, technical reports, datasets, and open-source releases.