NeuroForge
Choose appearance
Appearance

Use your device setting or keep a preference on this browser.

Work with us
Navigate
See the 10-Day AI Fit TestChoose a path

Quick find

Navigate NeuroForge

Esc

Start typing to search public pages and evidence.

    ERAIS listening

    The new speech set is ready for your ears.

    Four qualified concealed pairs are now open for product feedback. Listen to both versions, rate their quality, choose a preference, and send one privacy-minimised response.

    Start the four-pair comparisonRead the retained V4 failureTry the browser instrument

    Open now

    Speech K7 concealed comparison

    A deliberately small four-pair task built from the qualified natural-timing run.

    Audio qualifiedControl concealed

    LuxTTS control vs ERAIS K7

    Judge naturalness, clarity, timing, and unwanted artifacts. Both routes transcribed every word correctly on this exact set; your ears decide whether one is actually preferable.

    • 8 × 48 kHz speech clips
    • 4 concealed A/B pairs
    • About 5 minutes
    • Headphones recommended

    Exactly eight ratings and four preferences. No free text or identity profile.

    The collection closes at 500 complete responses and automatically removes records after 30 days.

    Listen and rate
    Failure measuredCollection closed

    Audio generation V4 · retained failure

    The earlier 24-file set remains offline after objective screening exposed unacceptable non-speech outputs. Its failure receipt stays public so the bad result does not disappear.

    The failed WAVs, players, and response flow remain quarantined.

    Inspect the retained failure

    How it works

    One careful pass. No identity profile.

    1. 01

      Prepare

      Be 18 or older, find a quiet place, use a comfortable volume, and read the short consent information.

    2. 02

      Listen with roles hidden

      Finish A and B in each pair. The page never tells you which route made either clip.

    3. 03

      Make 12 decisions

      Rate all eight clips from 1–5 and make four A/B/tie choices. There is no inferred default.

    4. 04

      Submit once

      One complete pseudonymous response is stored after the anti-abuse check. Close the page before then to stop.

    Release boundary

    Product feedback is not research evidence.

    The K7 release binds the participant task to its exact WAV identities, qualification receipt, operator review, and rights receipt. Accepted responses remain a voluntary convenience sample and cannot establish population preference, scientific MOS, model quality, superiority, or state-of-the-art capability.

    Formal listening research remains a separate, held governance path. A formal study will appear here only after its publication rights, participant information, research governance, collection controls, and evidence contract are ready.

    Continue exploring