Machine Learning Researcher

Company Neuphonic is building the future of on-device voice AI. We develop ultra-low-latency voice models that enable super-realistic, human-like speech directly on devices. Our focus is on building efficient generative audio models and infrastructure that can run on CPU-constrained hardware, enabling real-time voice interaction without relying on large cloud infrastructure. By dramatically reducing latency and compute requirements, we're making natural conversational AI possible on phones, embedded devices, browsers, and edge systems. This requires building high-performance systems that bridge machine learning, real-time audio processing, and distributed infrastructure.

Neuphonic was founded in April 2024 and has raised significant funding from leading European investors. We work with global enterprises across hardware and consumer AI that are building the next generation of voice-enabled products. Our open source models have been downloaded over 4 million times and earned over 6,500 GitHub stars since launch. Our vision is a world where voice becomes the most natural interface for AI, enabling seamless, intuitive interactions that are accessible to everyone. Please review our Hugging Face and GitHub repos before applying, as they’ll be part of the interview discussion.

We are looking for a Machine Learning Engineer to help advance the state of the art in speech synthesis. You will work on research and development across the full speech pipeline — from model architecture and training to dataset design and production deployment. The role combines applied research with real-world engineering, working closely with a small team pushing the boundaries of real-time speech systems. We are particularly interested in candidates with experience in text-to-speech systems, or multimodal machine learning involving speech and audio.

Your work will include:

• Researching and developing state-of-the-art speech synthesis models
• Training and optimising models for high-quality, low-latency speech generation
• Building and curating high-quality proprietary speech datasets
• Improving model quality, expressiveness, and latency
• Working closely with engineers to bring research models into production systems
• Exploring multimodal approaches to speech and conversational AI

This role is best suited to candidates who have worked on research-grade machine learning models, rather than purely application-level ML systems.

You have:

• An MSc or PhD in machine learning, speech processing, computer science, or a closely related field
• Strong experience training and evaluating deep learning models using frameworks such as PyTorch, JAX, or TensorFlow
• Several years of research or industry experience developing machine learning models (this is not a graduate or entry-level role)

In addition, you should have experience in one of the following areas:

• Text-to-speech (TTS) or speech synthesis, including model architecture, training, or evaluation
• Multimodal machine learning involving audio, such as models combining speech, text, or audio modalities
• Experience working in research-oriented ML environments, such as academic labs, advanced research teams, or deep-tech startups
• Familiarity with state-of-the-art approaches in speech generation, audio modelling, or multimodal systems
• Experience reading and implementing recent ML research papers
• Background from top universities, leading research groups, or equivalent research experience
• A strong interest in speech technology and conversational AI

Benefits:

• Help us set the company culture - you’d be joining us as part of our founding team.
• Competitive salary and equity - we want to make sure you’re rewarded not just in financial terms but also have a sizeable stake in the business as well.
• Private health insurance - your health and well-being is paramount.
• Conference, travel and development budget - your development will also grow the business.