---
title: "Research — how Dialogist voice agents work"
description: "Research notes from the Dialogist team: latency, turn-taking, speech recognition and synthesis, measured on real calls and explained step by step."
canonical: "https://dialogist.cc/en/research/"
locale: "en"
updatedAt: "2026-11-11"
---

# Measurements from inside a phone call with a voice agent.

Latency, turn-taking, recognition, synthesis and testing, measured on real calls and at load, explained step by step.

## [Voicecore: Rust media backend, road to infinite scalability with no idle GPUs](https://dialogist.cc/en/research/voicecore-rust-media-core/)

2026-11-11 · Research, Capacity, Load

The Python media path used up the processor of one bot process at about 96 conversations. Voicecore, written in Rust, carries 224 with the GPU out of the way.

## [One Server Unit: 100 simultaneous calls, load-tested to 176% of rated capacity](https://dialogist.cc/en/research/server-unit-under-load/)

2026-10-28 · Research, Load, Capacity

We loaded one Server Unit from a single call to 176% of its rated 100 calls. Replies held the Standard grade up to rated capacity, and every call connected.

## [Oasis Speech 2: recognition in 48 languages, before and after](https://dialogist.cc/en/research/oasis-speech-2/)

2026-10-26 · Research, Speech recognition, Languages

Oasis Speech 2 lowers the error rate on short replies by 44% across the 48 languages it recognizes, and all 48 improve.

## [Listening while talking, and what the agent does with the caller's open channel](https://dialogist.cc/en/research/listening-while-talking/)

2026-10-24 · Research, Turn-taking

How a Dialogist agent keeps listening while it speaks, starts its reply inside the pause and tells an interruption from a cough.

## [The shortest answers are the hardest to recognize, and the most common on a call](https://dialogist.cc/en/research/short-replies/)

2026-10-22 · Research, Speech recognition

Most of what a caller says is a few words, and a few words are the hardest case for speech recognition. Oasis Speech 2 cuts errors on one-word replies by 49%.

## [How we grade a conversation between a voice agent and a simulated caller](https://dialogist.cc/en/research/how-we-grade-a-conversation/)

2026-10-21 · Research, Evaluation, Testing

Polygon, our test bench, grades a voice agent on whole conversations, with simulated callers, eight weighted criteria, hard gates and behavior probes.

## [Cadence: the workbench where we fine-tune, evaluate and gate the models behind Oasis Speech](https://dialogist.cc/en/research/cadence/)

2026-10-20 · Research, Speech recognition, Training

Cadence is where we fine-tune and evaluate the models behind Oasis Speech. People and coding agents use one API, and every model must pass a gate.

## [What happens to a sentence after the voice model speaks it](https://dialogist.cc/en/research/after-the-voice-model/)

2026-10-19 · Research, Speech synthesis, Latency

Between the voice model and the caller, the agent trims, resamples, checks and caches every sentence, and re-checks the cache at night.

## [Knowing when the caller has finished](https://dialogist.cc/en/research/knowing-when-the-caller-has-finished/)

2026-10-14 · Research, Turn-taking

Across nearly 100,000 caller turns, our end-of-turn model closed most turns before the silence timer and gave back about 0.3 s per turn on real calls.

---

Contact: hello@dialogist.cc