Machine Learning TTS Engineer

We’re expanding our TTS to new languages and moving to next-generation LLM-based TTS, and we’re looking for a Machine Learning TTS Engineer to help bring these technologies to production quality! You’ll join a small TTS team of 2–3 ML engineers, supported by a dedicated MLOps team and a data engineer. Requirements: - 3+ years of experience in speech technology with hands-on TTS experience. - Hands-on experience with modern LLM-based TTS approaches (e.g., VALL-E, CosyVoice, XTTS). - Experience making TTS models production-ready: stability, robustness, and inference efficiency at the model level. - Proficiency in deep learning frameworks, especially PyTorch. - Understanding of modern ML fundamentals: Transformers, diffusion/flow matching, neural audio codecs, and vocoders. - Understanding of streaming TTS constraints (latency, chunked/incremental generation) and how they shape model architecture choices. - Knowledge of audio signal processing and the linguistic side of TTS (phonetics, text normalization). - Basic experience with speech data processing (segmentation, alignment, quality filtering). - Familiarity with TTS evaluation methods (MOS, SBS). - Strong Python skills and solid software engineering practices. - Proven ability to independently explore new ideas and improve existing ML systems. - BS or higher in Computer Science or a related field, or equivalent experience. - English B2+. Nice to have: - Experience with low-resource languages, including dataset creation and augmentation. - Experience designing recording scripts and specs for TTS voice datasets. - MLOps/DataOps experience, e.g., MLflow, DVC, or similar experiment tracking tools. - C++ skills. - Publications in speech/ML venues. Responsibilities: Extend our TTS to new languages: data collection and curation, text normalization/G2P, and model adaptation. Take new TTS architectures from research to production: improve robustness, latency, and quality, working closely with MLOps on deployment. Push existing models further through post-training, data curation, architectural improvements, and new training approaches. Build efficient audio data pipelines, from speaker recording and preprocessing to model training. Design rigorous benchmarks and evaluations that show whether a new model iteration is genuinely better, and why. What we offer: - The team has built award-winning AI products for tech corporations — devices, voice assistants, products that are actually in the world - Cutting-edge tech stack: Speech Technologies, NLP, Generative AI (LLMs, diffusion models), voice-first agentic architecture with privacy-first and on-premises deployment - High engineering bar and real ownership — the team cares about what actually works in production, not what looks good in a demo, and you'll see the impact of your work directly - Fast career progression — a senior-heavy team and a high volume of real problems means you grow faster than you would anywhere else - Startup pace with enterprise stability — real clients, real revenue, no bureaucracy - Fully remote across Europe - 21 vacation days + public holidays + 5 sick days - Private English lessons via Preply