protege - Machine Learning Researcher - Audio
Upload My Resume
Drop here or click to browse · Tap to choose · PDF, DOCX, DOC, RTF, TXT
Requirements
• 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. • 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. • 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.
Responsibilities
• 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. • 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.
No credit card. Takes 10 seconds.