Preview Audrey 0.7: speech recognition for real calls
Join the waitlist

Speech models for
real conversations.

Audrey turns speech into accurate text. turn-1 knows when the speaker has finished. Built for voice products that talk to people.

Join the waitlist Meet Audrey
turn-1-mini · Apache-2.0Sesame TurnBench · held-out test set
1stof 23End-of-turn on Sesame's TurnBench
2ndof 20Interruption on Sesame's TurnBench
5.3MParameters. 5.4 MB with 8-bit weights
1.3msPer 160 ms step on one CPU core (Apple M4 Pro)
Plug-ins for Pipecat and LiveKitGet the weights
Products

Two models, one conversation

In preview

Audrey 0.7

Speech recognition for real calls.

Accurate text from accented speakers, crosstalk and noisy rooms, including the names and numbers your workflow depends on.

Request access Learn more
Open weights

turn-1

Turn detection for voice agents.

Knows when the speaker has finished, so your agent neither cuts people off nor leaves them waiting.

Get the weights Learn more
Blog

Blog

All posts
Results

turn-1-mini on Sesame's TurnBench

Ranked first on end-of-turn and second on interruption on the held-out test set. What the benchmark measures, and the full numbers with latency spread.

Kollaikal RupeshRead →
Explainer

What a turn detector decides at every pause

Finished, or just pausing? The one decision our models make, the two ways to get it wrong, and where it sits in a voice agent.

Kollaikal RupeshRead →
Technical note

turn-1 and turn-1-mini

Our two end-of-turn models: sizes, speed, how they are built, and their licences.

Kollaikal RupeshRead →
Waitlist requestReply by email

We use your email only to contact you about p99lab. See the privacy policy.

Waitlist

Join the waitlist

Building a voice product? Join the waitlist for early access to Audrey and to new p99lab models. We will email you when there is something to try.