Machine Learning Researcher - Audio

Responsibilities

• Research audio data quality for machine learning by investigating how signal properties and dataset composition affect downstream model training.
• Develop new metrics, benchmarks, and evaluation frameworks to measure audio quality in ways that predict ML model performance.
• Characterize speech datasets by analyzing acoustic properties such as effective bandwidth, spectral energy, noise, and codec artifacts.
• Build workflows for segment-level quality evaluation to detect localized degradation in diarized or segmented speech regions.
• Design and run targeted evaluations connecting audio quality issues to downstream behaviors in ASR, TTS, and speaker modeling.
• Translate research findings into reproducible filtering rules, quality gates, and scalable evaluation infrastructure.
• Collaborate with ML researchers, data engineers, and operations teams to communicate the value of audio data assets.

Requirements

• PhD or equivalent Master’s degree plus 4+ years of industry experience in machine learning, audio signal processing, or speech technology.
• Proven experience designing and running data evaluations, audio analyses, or benchmarks.
• Strong understanding of speech/audio signal properties, including sampling rates, codecs, spectrograms, and perceptual quality.
• Experience developing or evaluating metrics and measurement frameworks for ML systems or audio signal analysis.
• Ability to connect low-level signal properties to downstream machine learning behavior and model robustness.
• Proficiency in moving between research exploration and production implementation of scalable tools.
• Excellent technical communication skills and a high degree of ownership.

Preferred Qualifications

• Experience with ASR, TTS, speaker modeling, self-supervised speech models, or multimodal audio models.
• Experience developing evaluation frameworks specifically for training data.
• Publications or open-source contributions in speech, audio ML, or data-centric AI.
• Experience studying the relationship between dataset quality and downstream model performance.

About the Company

Protege is building a platform to solve the biggest unmet need in AI: access to high-quality training data. We facilitate the secure, efficient, and privacy-centric exchange of AI training data, helping ambitious teams power their models with the best possible signals. We are a lean, fast-moving, high-trust team of builders obsessed with velocity and impact.