Version 0.2-draft · Effective

RemoteDataWork Acceptance Criteria

Version 0.2-draft

Draft for counsel review. Not in force. This draft describes how recordings would be accepted or rejected. It has not been approved, and no recording is reviewed under it until a final version is published here with its effective date.

These Acceptance Criteria explain how Speechify, Inc., which operates RemoteDataWork, decides whether a recording of a conversation session is accepted. Only accepted recordings are paid at the project's rate per accepted recorded hour. The qualification session is the exception: its flat fee is paid whether or not it is accepted.

The criteria are about whether a recording is usable and was made under the Program's rules. They do not grade how interesting, fluent or skilled the conversation is, and they never rank contributors.

1. What is reviewed

  • Each recording is reviewed per contributor: your side of the session is accepted or rejected on its own, so another participant's problem doesn't cost you your pay.
  • Each contributor's voice is recorded on its own track. Review looks at your track and at what can be heard in it.
  • A person reviews every recording. Automated checks may flag audio problems for that person to look at, but a rejection is always decided by a person and always comes with one of the reasons in Section 3.

2. What an accepted recording looks like

A recording is accepted when all of the following are true:

  1. It's a live conversation in your own voice. You speak live, through the Platform, with the other enrolled participants in the session. No playback, prerecorded, synthetic or altered audio, and no one speaking for you.
  2. It's natural speech. You talk freely about the topic. Reading a text aloud, reciting from a script or repeating set phrases is not natural speech.
  3. Only enrolled participants can be heard. No one else, including children, is audible in your track: no other person in the room, no call on speaker, no television or radio with speech, and no music playing in the background.
  4. It follows the Conversation Guidelines, including the privacy, copyright and Conduct and Safety rules.
  5. The audio is clear enough to use. Your speech can be heard and understood throughout. In particular:
    • your voice is clearly above the background (not so quiet that it is lost in room noise);
    • your voice is not distorted by clipping (too loud for the microphone);
    • there is no constant hum, buzz or hiss loud enough to cover speech;
    • there are no long gaps or dropouts where the audio cuts out.
  6. It's complete. You took part in the conversation for the length of the session you are paid for, and your track contains your side of it.
  7. It was recorded under the Program's rules. Every participant tapped to start before recording began, you had signed the documents in force for that session, and you joined from an eligible location.

3. Reasons a recording may be rejected

Every rejection gives one of these reasons, shown on your account page and in the decision email.

Reason What it means Why it matters
Audio quality Noise, clipping, hum, hiss or dropouts make your speech hard to hear or understand for much of the recording. Speech AI can't learn from speech it can't hear clearly.
Third party audible Someone who isn't an enrolled participant in the session can be heard in your track. That person hasn't consented to being recorded, and their voice can't be used.
Not natural speech The speech is read aloud, scripted, or otherwise not a natural conversation. The Program collects natural conversation; read speech is a different kind of data.
Guidelines The recording breaks the Conversation Guidelines, for example by sharing someone's personal details, reading copyrighted text aloud, or breaking the Conduct and Safety rules. These rules protect other people, other participants and you.
Incomplete Your track is too short to use, or you left the session long before it ended while being paid for its full length. Pay is for the recorded time you took part in.
Technical failure A technical problem made your track unusable, for example it is missing, empty or corrupted. There is nothing usable to accept. Where the problem was on the Platform's side, we say so in the reason.

A recording made by fraud, such as playback or synthetic audio, someone else speaking on your account, or falsified session time, is treated as rejected under Section 7.3 of the Contributor Services Agreement.

4. How much of a recording is paid

  • Accepted recorded time is the recorded length of the session you took part in, from when recording started to when it ended, for each recording that is accepted.
  • If part of a recording is unusable but the rest is fine, the reviewer may accept it for the usable part only. The decision then shows the time accepted and the reason the rest wasn't.
  • Pay is prorated by the minute at the rate shown when you accepted the project.

5. Review and appeal

  • We review each recording within 10 business days after the session.
  • If a recording is rejected, your account page shows the reason. You can appeal from your account within 14 days.
  • A different reviewer decides the appeal within 10 business days. If the appeal succeeds, the recording is accepted and paid on the next pay day. If not, you're told why.
  • Rejected recordings are never used to train AI models or included in any dataset. They are deleted within 30 days after the appeal period ends, except while needed to decide an appeal or look into a possible breach of the Conversation Guidelines.

6. Changes

We may update these criteria. The version in force and its effective date are shown at the top of this page, and a recording is reviewed under the version in force when it was made.

Questions

Questions about a decision go to support@remotedatawork.com, or use the appeal on your account page.