Staff AI Speech Engineer — Real-Time Voice & Telephony

Salient; United States; Remote; USD 180,000 - 280,000; Full time

Benefits offered by this job

Medical, dental, and vision coverage

401(k)

Catered lunches

Equity

Job summary

Salient is hiring to design, train, and ship speech-to-speech and real-time voice models for live calls, building a robust production speech stack with neural codecs, streaming inference, and turn detection. You will develop evals and monitoring for real borrower conversations and diagnose challenges like telephony noise and barge-in in end-to-end systems.

You will own major technical areas from problem definition to production deployment, collaborating across models, data, infrastructure, and

Qualifications

• 5+ years in speech, audio, or related ML, or demonstrated depth through shipped systems.
• Proven work on speech-to-speech, full-duplex, or real-time conversational speech systems.
• Public proof of work such as papers, GitHub repositories, or Hugging Face models.
• Strong engineering fundamentals and reliable low-latency production systems.
• Ability to diagnose messy real-world model failures.
• High ownership and comfort in ambiguous technical areas.

Responsibilities

• Design, train, and ship speech-to-speech and real-time voice models for live calls.
• Build and improve the production speech stack: neural codecs, streaming inference, turn detection, speech enhancement.
• Develop evals and monitoring reflecting real borrower conversations.
• Diagnose real-world failures such as telephony noise, barge-in, and overlapping speech.
• Work across models, data, infrastructure, and product to raise end-to-end agent quality.
• Own major technical areas from problem definition to production deployment.

Skills

Speech ML

Real-time systems

Low-latency infra

Model failure diagnosis

Production deployment

Job description

Salient is hiring to design, train, and ship speech-to-speech and real-time voice models for live calls, building a robust production speech stack with neural codecs, streaming inference, and turn detection. You will develop evals and monitoring for real borrower conversations and diagnose challenges like telephony noise and barge-in in end-to-end systems.

You will own major technical areas from problem definition to production deployment, collaborating across models, data, infrastructure, and