What the work involves

You'll record voice samples across a spread of script types — conversational exchanges, instructional prompts, short narrative passages, and corporate-toned service lines — that will be used to train and evaluate a text-to-speech system powering a client's internal CX agent. Sessions are self-directed: you receive scripts and a recording spec, record in your own space, and deliver files that meet formatting and noise-floor requirements. Expect to do multiple takes of the same line with deliberate variation in emotion, emphasis, and pace, and to return across weeks with your accent, mic distance, and vocal placement matched to earlier sessions.

The voice cloning term is central, not incidental. The client intends to build a synthetic voice from your recordings for their own CX agent; the listing states the recordings will not be sold, licensed, or reused for other products or datasets. Read the agreement carefully and only apply if you're comfortable with that arrangement.

What the screen looks for

  • Accent authenticity and provenance. Yorkshire, Manchester, Newcastle, Liverpool and neighbouring varieties — natural, not performed, and you should currently be based in the northern UK.
  • Verifiable professional history. Voice acting, dubbing, narration, podcasting, or broadcast credits you can name and point to. TTS, audiobook, IVR, or AI dataset experience is a plus.
  • Technical setup. A specific microphone, interface, and treated or quiet room — expect to name your gear and describe your noise floor rather than assert it's "professional."
  • Consistency discipline. Evidence you can reproduce a delivery weeks later, and that you understand why drift matters for a cloned voice.

Logistics

Fully remote and asynchronous. Roughly 5–10 hours per week for the project duration, scheduled around your own availability, with turnaround windows on batches rather than fixed session times. Deliverables are audio files to a stated spec; light self-editing familiarity (Audacity, Audition, Reaper) helps but heavy post-production is generally discouraged for training data. Pay is listed at $50/hr as observed on the platform, not guaranteed — rates and hours can shift with project scope.