Apply on Employer Site

OutcomesAI · 2 months ago

Tech Lead — ASR / TTS / Speech LLM (IC + Mentor)

Boston, MA

Full-time

Hybrid

Senior Level, Lead/Staff

7+ years exp

OutcomesAI is a healthcare technology company focused on building an AI-enabled nursing platform to enhance clinical workflows and patient care. The Tech Lead will oversee the technical development of speech models, guiding a small team in model training and optimization for healthcare applications.

Artificial Intelligence (AI)Health CareMedical

Responsibilities

Own the technical roadmap for STT/TTS/Speech LLM model training: from model selection → fine-tuning → deployment

Evaluate and benchmark open-source models (Parakeet, Whisper, etc.) using internal test sets for WER, latency, and entity accuracy

Design and review data pipelines for synthetic and real data generation (text selection, speaker selection. voice synthesis, noise/distortion augmentation)

Architect and optimize training recipes (LoRA/adapters, RNN-T, multi-objective CTC + MWER)

Lead integration with Triton Inference Server (TensorRT/FP16) and ensure K8s autoscaling for 1000+ concurrent streams

Implement Language Model biasing APIs, WFST grammars, and context biasing for domain accuracy

Guide evaluation cycles, drift monitoring, and model switcher/failover strategies

Mentor engineers on data curation, fine-tuning, and model serving best practices

Collaborate with backend/ML-ops for production readiness, observability, and health metrics

Qualification

Speech models expertisePyTorchTriton Inference ServerKubernetesStreaming RNN-TTelephony robustnessSpeaker DiarizationTTS frameworksSpeech LLM integrationCode reviewMentorship

Required

M.S. / Ph.D. in Computer Science, Speech Processing, or related field

7–10 years of experience in applied ML, at least 3 in speech or multimodal AI

Track record of shipping production ASR/TTS models or inference systems at scale

Deep expertise in speech models (ASR, TTS, Speech LLM) and training frameworks (PyTorch, NeMo, ESPnet, Fairseq)

Proven experience with streaming RNN-T / CTC architectures, LoRA/adapters, and TensorRT optimization

Telephony robustness: Codec augmentation (G.711 μ-law, Opus, packet loss/jitter), AGC/loudness norm, band-limit (300–3400 Hz), far-field/noise simulation

Strong understanding of telephony noise, codecs, and real-world audio variability

Experience in Speaker Diarization, turn detection model, smart voice activity detectionEvaluation: WER/latency curves, Entity-F1 (names/DOB/meds), confidence metrics

TTS : VITS/FastPitch/Glow-TTS/Grad-TTS/StyleTTS2, CosyVoice/NaturalSpeech-3 style transfer, BigVGAN/UnivNet vocoders, zero-shot cloning

Speech LLM: Model development and integration with Voice agent pipeline

Experience deploying models with Triton Inference Server, Kubernetes, and GPU scaling

Hands-on with evaluation metrics (WER, F1 on entities, latency p50/p95)

Familiarity with LM biasing, WFST grammars, and context injection

Strong mentorship and code-review discipline