Kaldi Production Engineer Jobs

Why Kaldi Production Engineering?

Kaldi remains the production workhorse for enterprise ASR systems despite the rise of Whisper and other deep learning models. Companies processing millions of hours annually rely on Kaldi's proven stability, CPU efficiency, and real-time streaming capabilities.

What You'll Work On

• Real-time ASR systems: Streaming speech recognition with <100ms latency for call centers, voice assistants, and live captioning
• Large-scale deployments: Systems processing 100K+ hours/day with 99.9% uptime requirements
• Model optimization: Reducing WER through acoustic model tuning, language model adaptation, and pronunciation optimization
• Production infrastructure: Docker/Kubernetes deployments, load balancing, auto-scaling, monitoring
• Custom recipes: Building domain-specific Kaldi recipes for healthcare, legal, finance, and other verticals

Key Technologies

• Kaldi toolkit: nnet3, chain models, online decoding, lattice rescoring
• C++/Python: Kaldi core (C++) with Python wrappers for services
• GPU acceleration: CUDA for acoustic model inference, TensorRT optimization
• Streaming protocols: WebSocket, gRPC, RTMP for real-time audio
• Infrastructure: Docker, Kubernetes, Terraform, AWS/GCP

Who's Hiring

Companies maintaining large-scale Kaldi deployments include:

• Big Tech: Apple (Siri backend), Amazon (legacy Alexa systems), Microsoft (Teams transcription)
• Contact Centers: CallMiner, Verint, Nice (billions of customer calls)
• Speech Analytics: Gong, Chorus (migrating but still using Kaldi for streaming)
• Telecom: AT&T, Verizon (voicemail transcription, call analytics)
• Government/Defense: Classified ASR systems requiring on-premise deployment

Salary Ranges by Experience

Companies hiring for Kaldi / production-ASR roles

These companies build in this space and hire for Kaldi / production-ASR work. Specific openings come and go — check their careers pages, and get the current ones each week in the SpeechTechJobs digest.

• CallMiner — conversation analytics for contact centers
• Dialpad — an AI cloud phone and contact-center platform
• Verint — customer-engagement and workforce-optimization software
• Rev — transcription and captioning at scale
• Speechmatics — a speech-to-text API company

Skills & Requirements

Core Technical Skills

• Kaldi expertise: nnet3 recipes, chain models, online decoding, lattice rescoring
• Programming: C++ (Kaldi internals), Python (services, automation)
• Linux/Unix: Shell scripting, system administration, performance tuning
• Real-time systems: WebSocket, gRPC, low-latency optimization
• DevOps: Docker, Kubernetes, CI/CD, monitoring (Prometheus, Grafana)

Desirable Skills

• GPU optimization (CUDA, TensorRT)
• Experience with other ASR frameworks (Whisper, Wav2Vec 2.0)
• Acoustic modeling (DNN, TDNN, LSTM)
• Language modeling (n-gram, neural LMs)
• Cloud platforms (AWS SageMaker, GCP AI Platform)

Typical Interview Process

1. Phone Screen (30 min): Background, Kaldi experience, availability
2. Technical Screen (60 min): Kaldi architecture questions, debugging scenarios
3. Coding Challenge: Build Kaldi recipe or optimize existing pipeline
4. Onsite/Virtual (4-5 hours):
   • System design: Large-scale ASR architecture
   • Deep dive: Kaldi internals, optimization strategies
   • Coding: C++ or Python implementation
   • Behavioral: Past projects, team collaboration

Related Resources

• What is Kaldi? Complete Guide (2026) - Deep dive into the toolkit
• ASR Research Jobs - For Kaldi development and research roles
• Whisper vs Kaldi: When to Use Each
• Speech Tech Conferences 2026 - Network with Kaldi community