Home › Home studio requirements
What your recording space actually needs
Speech data is judged differently from voiceover. A room that sounds fine on a corporate read can be unusable for training audio, because whatever is in the background gets learned along with your voice. Here is the honest standard.
The room matters more than the microphone
Most rejected applications fail on the room, not the kit. A good microphone in a hard, echoey space produces audio that cannot be fixed later. A modest microphone in a soft, quiet room usually produces usable data.
- Consistent quiet. No traffic peaks, no boiler cycling, no fridge compressor, no extractor fan, no neighbours’ music. Consistency across a long session is the test, not one quiet minute.
- Soft surfaces. Curtains, carpet, a wardrobe of clothes, a duvet behind you. Bare walls and laminate floors put reflections into every take.
- No electrical hum. Mains buzz, dimmer switches and cheap USB power are common and hard to remove cleanly.
- Nothing that changes. Same room, same position, same distance, same gain, for the whole project. Dataset audio has to match itself from the first file to the last.
Technical spec we normally ask for
- Format: uncompressed WAV, mono.
- Sample rate and depth: 48kHz / 24-bit or better. Never a lossy format such as MP3 or AAC.
- Processing: none. No noise reduction, no compression, no EQ, no gates. Raw audio is what we need - processing destroys information the model needs and cannot be undone.
- Headroom: record with peaks comfortably short of clipping. Quiet-but-clean is recoverable. Clipped is not.
The “no processing” rule surprises people who come from commercial voiceover, where a polished delivered file is the product. In dataset work the raw file is the product.
No studio? Still apply
A significant share of our work is recorded in controlled facilities rather than at home, particularly voice cloning sessions and anything where acoustic consistency has to be guaranteed. If you have a strong or under-recorded accent, or a language we are short of, apply anyway and say honestly what your setup is. Overstating your kit only wastes a session later.
Not sure whether the work suits you? Read what a session actually involves or how to start with no experience. Or see what is casting now, and how to sell your voice to AI on fair terms.
Kit and room questions
Do I need a home studio for AI voice work?
Not always. Some projects are recorded in our studios or a partner facility, which removes the requirement entirely. For remote sessions you need a consistently quiet room and a microphone that is not your laptop’s built-in one. A treated professional booth is a bonus, not an entry requirement.
What microphone do I need?
Any decent USB or XLR condenser or dynamic microphone on a stand, recording at 48kHz/24-bit or better, mono, uncompressed WAV. What matters more than the model is that it stays in the same position for the whole project - consistency beats specification in dataset work.
What noise floor do I need to hit?
As a working guide, a room quiet enough that a recorded silence sits well below the speech and carries no hum, hiss, traffic or fan noise. Speech data has less tolerance for background noise than a commercial voiceover, because the noise becomes part of what the model learns.
Can I record on my phone?
For the application, yes - a voice note is enough for us to hear your voice and accent. For paid dataset sessions, no. Phone recordings are compressed and processed in ways that make them unusable as training audio.
Tell us honestly what you have
We would rather know your real setup than discover it mid-session.
Join the roster
Apply as voice talent
Two minutes, no demo reel required - a link to anything we can listen to helps, but is not essential for dataset work. We review every application personally and only contact you about real, paid projects.