Job Title: Senior Speech & Audio ML Engineer Location: Remote Type: Full-time Salary Range: $2,500–5,500 USD / month Role Purpose We are looking for a Senior ML Engineer to build and ship core models for a speech-driven behavioral engine. You will own end-to-end modeling from raw, long-form audio and layered annotations to production inference. Responsibilities include: Designing audio features and embeddings. Training and evaluating a suite of models. Delivering reproducible pipelines that meet targets for accuracy, robustness, latency, and cost. Non-Negotiables Experience: 5+ years building production ML systems, including 2+ years in speech/audio. Speech & Signal Processing: VAD, diarization, segmentation, denoising, spectral features (log-mel/MFCC), prosody (pitch/energy), long-form audio handling. SOTA Audio Models & Embeddings: Wav2Vec2, HuBERT, wavLM (or similar); fine-tuning/self-supervised learning; contrastive/metric learning for downstream tasks. Data Engineering & Quality: SQL, Python data stack (Pandas/Polars), ETL for audio + metadata, stratified sampling, leakage prevention, feature stores. Evaluation Discipline: Golden sets, robust speaker/content splits, ROC/PR/calibration, fairness/bias checks, ablations, drift/shift detection on embeddings and audio quality. MLOps, Serving & Reproducibility: FastAPI/gRPC around HF/torchaudio models, experiment tracking (W&B/MLflow), artifact/model versioning, CI/CD, observability, scalable batch/streaming inference. Proven ability to create and document novel IP (methods, architectures, or training/eval techniques) with clear prior-art awareness. Nice to Have Tooling: SpeechBrain, Lightning, OpenSMILE/Praat, Kaldi/Conformer/Emformer, Label Studio. Multimodal Skills: ASR (e.g., Whisper) + paralinguistic features; emotion/prosody modeling; speaker embeddings (x-vectors, ECAPA-TDNN). Performance & Deployment: Quantization/distillation, Triton/CUDA basics, distributed training, real-time/streaming inference, on-device DSP (Rust/C++). Publications/Patents/Competitions: Demonstrating novel audio modeling work.