METHODOLOGY
How the Listening Lab works
What this is
The Speko Listening Lab is a paid listening panel. Speko runs a public voice benchmark at arena.speko.ai, where anyone can vote in blind A/B listening tests. The Listening Lab is a controlled companion to that benchmark: partner research panels recruit and compensate listeners, and those listeners complete structured sessions here. We use these sessions to calibrate the public benchmark - to check that an open, self-selected vote pool and a paid, quality-controlled panel point in the same direction.
How a session works
Each session is a fixed series of short tasks. In every task you see one short text and hear two recordings of it, labeled Voice 1 and Voice 2. You pick the one that sounds more natural. The choice is forced: there is no tie and no skip. Which recording plays as Voice 1 or Voice 2 is randomized for every task, and the identities behind the recordings are never shown.
Quality control
Sessions include attention checks. We do not disclose how many there are or what they look like. We also require that both recordings are actually played before a choice can be made, and we measure listening and response times. Responses that fail these checks are excluded from analysis. Whether a completed session passes quality control is reported to the partner platform that recruited the listener; compensation is handled entirely by that platform.
A separate pool
Votes collected in the Listening Lab are stored separately from the organic votes cast at arena.speko.ai. The two pools are never mixed. Any published analysis that uses Listening Lab data labels it as paid panel data.
Data handling
Sessions are anonymous. We store the participant code provided by the partner platform (so the partner can verify completion), a random browser identifier used to prevent duplicate sessions, the choices made, and technical signals such as listening time. We do not collect names, email addresses, or any other personal information. Results are used only in aggregate.