archive

The audio engineering behind low-latency voice systems. After working through LiveKit and Pipecat internals, I wanted to understand what really happens once TTS audio reaches the transport layer. This post is about how audio gets framed and paced, how VAD knows when you start and stop talking, how bursty TTS output becomes steady, click-free speech in real time, and how jitter buffers and backpressure keep the pipeline under control.

Sample rates are not quality settings. They are engineering decisions shaped by whether a system is reproducing music, understanding speech, or carrying a phone call as cheaply as possible.

A breakdown of Raft: how distributed servers elect a leader, replicate logs, and stay consistent even when nodes fail.

Every retrieval algorithm exists because the previous one had a flaw. This is a deep dive into how search evolved, and why it had to.

A hands-on walkthrough of building a real-time voice agent with LiveKit, wiring up the full STT to LLM to TTS loop so you can have back-and-forth voice conversations.

A breakdown of how real-time voice agents work under the hood: the two-layer architecture, VAD, STT, LLM, TTS pipeline, and where latency comes from at each stage.