Company Overview:
We are building Protege to solve the biggest unmet need in AI — getting access to the right training data. The process today is time intensive, incredibly expensive, and often ends in failure. The Protege platform facilitates the secure, efficient, and privacy-centric exchange of AI training data.
Solving AI’s data problem is a generational opportunity. We’re backed by world-class investors and already powering partnerships with some of the most ambitious teams in AI. The company that succeeds will be one of the largest in AI — and in tech.
We’re a lean, fast-moving, high-trust team of builders who are obsessed with velocity and impact. Our culture is built for people who thrive on ambiguity, own outcomes, and want to shape the future of data and AI.
Role Overview
Data is the foundation of AI performance, and we believe model quality starts with data quality. For speech and audio models in particular, the bar for signal fidelity, consistency, and quality control is exceptionally high.
We’re seeking a Machine Learning Researcher focused on audio data quality, ML data evaluation, and quality control to lead the evaluation and optimization of large-scale speech datasets used to train audio, speech, and multimodal models. This role will be responsible not only for applying existing audio quality metrics, but also for researching how audio data quality should be evaluated for machine learning systems and developing new methods, benchmarks, and evaluation frameworks that better predict downstream model performance.
You will help define what “high-quality audio data” means in the context of modern ML training. That includes studying how different forms of acoustic degradation, dataset inconsistency, recording conditions, speaker variation, labeling quality, segmentation quality, and signal artifacts affect model behavior across ASR, TTS, speaker modeling, representation learning, and multimodal systems.
A core part of this role will be original research and method development: designing new approaches for measuring audio data quality, validating those approaches against downstream model outcomes, and translating research insights into practical evaluation tools, filtering rules, and quality standards used across Protege’s data platform.
This is an ideal role for someone deeply obsessed with audio data quality and signal understanding, comfortable operating in both research and hands-on implementation modes, and excited to help Protege become the ubiquitous platform for high-quality AI training data.
What You’ll Do
Research audio data quality for machine learning
Investigate how audio quality, signal properties, dataset composition, and localized acoustic issues affect downstream model training, evaluation, and deployment.
Develop new metrics, benchmarks, diagnostics, and evaluation frameworks for measuring audio data quality in ways that are predictive of ML model performance.
Speech dataset characterization and metrics
Analyze and summarize Protege’s audio catalog and maintain clear, up-to-date quality scorecards and metrics for key speech datasets.
Develop methods to measure true acoustic properties directly from the waveform, including effective bandwidth, spectral energy distribution, high-frequency roll-off, noise, clipping, reverberation, distortion, and codec artifacts.
Segment-level quality evaluation
Build workflows that evaluate diarized or segmented speech regions, surfacing localized degradation that file-level averages may miss.
Apply multiple complementary quality metrics to detect bandwidth mismatches, resampling artifacts, clipping, reverberation, codec distortion, and other forms of degradation.
Model and data evaluation
Design and run targeted evaluations connecting audio quality issues to downstream model behavior, including ASR performance, speaker embedding stability, learned speech representations, and synthesis quality.
Test which audio quality metrics meaningfully correlate with model outcomes, identify failure modes of existing metrics, and design better alternatives when current approaches are insufficient.
Deterministic filtering and evaluation infrastructure
Translate research findings into reproducible filtering rules, quality gates, and dataset selection strategies that improve dataset consistency across training runs.
Build scalable tools and pipelines for applying audio quality analyses across large datasets, tracking results over time, and making quality signals accessible to researchers, engineers, and data teams.
Cross-functional collaboration
Work closely with ML researchers, data engineers, data operations, and external partners to define, measure, and communicate the value of Protege’s audio data assets.
What Success Looks Like
Near-term: establish a trustworthy audio-quality baseline
Create a trustworthy view of the quality, consistency, signal fidelity, and training-readiness of Protege’s speech and audio datasets, supported by metrics and scorecards the team can operationalize.
Then use targeted evaluations, ablations, and downstream model analysis to connect audio-quality issues to concrete dataset improvements and clearer prioritization over time.
What You Bring
PhD or equivalent Master’s degree + 4+ years industry experience in machine learning, audio signal processing, speech technology, computer science, statistics, engineering, or a related quantitative field.
Proven experience designing and running data evaluations, audio analyses, benchmarks, ablations, or slice-based analyses.
Strong understanding of speech/audio data and signal properties, including sampling rates, codecs, bandwidth, spectrograms, reverberation, clipping, noise, and perceptual quality.
Experience developing or critically evaluating metrics, benchmarks, or measurement frameworks for ML systems, data quality, speech technology, or audio signal analysis.
Ability to connect low-level signal properties to downstream machine learning behavior, including model accuracy, robustness, representation quality, speaker consistency, or synthesis quality.
Comfortable moving between research exploration and production implementation: you can formulate hypotheses, run experiments, analyze results, and turn findings into scalable tools or decision rules.
Excellent written and verbal communicator; able to write concise technical docs and explain empirical results clearly.
High ownership and bias toward action; you independently scope questions, design experiments, and drive them to decisions.
Bonus Signals
Experience with ASR, TTS, speaker modeling, self-supervised speech models, diarization, or multimodal audio models.
Experience developing evaluation frameworks or performance metrics for training data.
Experience inventing, adapting, or validating audio quality metrics for ML training datasets.
Experience studying the relationship between dataset quality and downstream model performance.
Publications or open-source contributions in speech, audio ML, data-centric AI, ML evaluation, or related areas.
Cross-functional collaboration with product, infrastructure, data operations, or partnership teams.
Experience collaborating with industry or academic labs on speech/audio research or data projects.
Protege Values
Pass the Loved Ones’ Test
We act with integrity and do the right thing — especially when it’s hard and no one is watching.
Always Find a Way
We are resourceful, resilient builders who solve hard problems and push through obstacles.
Go Fast and Grow Fast
Velocity matters. We move with urgency, learn quickly, and continuously improve as individuals and as a company.
Practice Kindness and Candor
We communicate directly and respectfully, building trust through honest feedback and genuine care for one another.
Deliver Together
We win as one team. Collaboration, accountability, and shared ownership drive our success.
Own the Outcome. Hone the Craft.
We take pride in our work, sweat the details, and continuously raise the bar for excellence.
