AI, data & engineering
Voice AI Engineer
Voice AI Engineer is an engineer who builds voice AI products: speech recognition, speech synthesis and voice agents that respond without noticeable delay.
Also known as: Speech Engineer, Speech AI Engineer, Voice Agent Engineer, Conversational AI Engineer, ASR Engineer, Speech Recognition Engineer.
What does a Voice AI Engineer do?
A Voice AI Engineer is an engineer who builds voice AI products: speech recognition (ASR), speech synthesis (TTS) and voice agents that talk to people in real time. The main constraint of the job is latency: the caller notices a pause, and before it becomes awkward the system has to recognise speech, get the model's answer and voice it. Typical tasks:
- build the ASR → LLM → TTS chain or work with speech-to-speech models, measuring latency at every step;
- solve the problems of live conversation: interruptions, pauses, end-of-turn detection, background noise;
- choose and fine-tune recognition and synthesis models for language, accents, terminology and the brand voice;
- connect telephony and WebRTC so the agent works in a call, in an app or in the browser;
- measure conversation quality: recognition errors, the share of conversations closed without a human, cost per minute.
When does a company need a Voice AI Engineer?
A company needs a Voice AI Engineer when voice is the product's main interface: voice agents for support and call centres, booking doctor appointments and phone sales, assistants in cars and devices, call transcription and analysis. A common trigger is a voice agent prototype built on off-the-shelf APIs that looks good in a demo but in real calls lags, interrupts the customer and gets names and addresses wrong. The role will not work if voice is a secondary feature and plugging in an off-the-shelf voice agent platform is enough: then an AI / LLM Engineer or an Agentic AI Engineer who configures its logic will do.
Voice AI Engineer vs AI / LLM Engineer: what's the difference?
A Voice AI Engineer differs from an AI / LLM Engineer in being responsible for a real-time conversation rather than a model's text answer: latency, audio, interruptions and telephony are their territory, and unfamiliar ground for an AI / LLM Engineer. What sets them apart from an NLP Engineer is the material: an NLP Engineer works with text, a Voice AI Engineer with sound and time.
Voice AI Engineer vs related roles: comparison
| Voice AI Engineer | AI / LLM Engineer | NLP Engineer | |
|---|---|---|---|
| Responsible for | A real-time voice conversation | An LLM product with a text interface | Language models for the task |
| Main result | An agent people find easy to talk to | A working LLM feature | A model more accurate or cheaper than an off-the-shelf one |
| Main constraint | Latency and audio quality | Answer quality and cost | Data and compute |
Why is it hard to hire a Voice AI Engineer?
Voice AI Engineers are hard to hire because the role is new and the experience it needs combines three areas that barely overlapped before: speech technology, LLMs and real-time systems.
- Speech specialists who worked before LLMs know ASR and TTS well but have not built agents; LLM product engineers, on the other hand, have not worked with audio and latency.
- Real-time experience — WebRTC, streaming, telephony — is more common among Backend Developers and VoIP engineers than among ML specialists.
- The title is unsettled: the same people call themselves Speech Engineer, Conversational AI Engineer or just AI Engineer.
What to decide before hiring a Voice AI Engineer?
- Are you building on an off-the-shelf platform or assembling your own chain? These need different people.
- What latency is acceptable? An agent for live calls and a service that transcribes recordings are different tasks, and without a target number it is hard to judge a candidate's experience.
- Do you need to train your own speech models? If so, you need ASR and TTS experience at the model level; if not, real-time engineering and integrations matter more.
Where to find a Voice AI Engineer?
Voice AI Engineers most often come from voice agent and call centre automation startups, speech technology companies (recognition, synthesis, transcription), voice assistant teams in devices and cars, and VoIP and communication platforms. A strong signal is a voice product with real calls or users and concrete latency numbers the person achieved. NEWHR finds such candidates by the results of their work, not by applications to a job post.
How to assess a Voice AI Engineer in an interview?
Voice AI Engineers are assessed with a review of a voice system the candidate launched and a case: design a voice agent for the company's task with a latency budget for each step. On the company side, the CTO or an ML Lead takes part.
Strong signals:
- breaks latency down by step in the chain and knows where the milliseconds go;
- raises interruptions, noise and end-of-turn detection without being asked;
- has listened to recordings of their agent's real conversations and analysed the errors.
Red flags:
- experience only with text chatbots, with voice "connected through an API";
- judges quality by the demo, without metrics on real calls;
- cannot say what latency their solution has.
How long does it take to hire a Voice AI Engineer?
We present the first candidates for the Voice AI Engineer role within 5–7 days of the brief.
Related roles
- AI / LLM Engineer — builds the agent's LLM logic, but usually not the audio, latency and telephony around it.
- Agentic AI Engineer — designs what the agent does in a multi-step task; the Voice AI Engineer makes it able to do that out loud without pauses.
- NLP Engineer — needed when the agent's weak spot is understanding language and terms rather than its voice.
- AI Infrastructure / Inference Engineer — cuts model serving latency at the GPU level, which a voice agent's latency budget depends on.