Prepare an Audio File for Noise Cleanup

Choose the right source copy

Use the earliest suitable recording you have rather than a repeatedly compressed messaging copy. Play it from beginning to end and confirm that the final sentence is present. Keep the original untouched. Renaming an extension does not convert a codec or repair a damaged file.

If several microphone recordings exist, compare them before choosing the processing source.

Check the actual requirements

Record the duration, size and intended output. Review your current plan and available minutes. Free Studio, paid Studio chunk uploads and direct API submission do not all have the same limits. Do not spend an allowance based on the assumption that all paid routes accept the same large file.

Remove unnecessary private material from a working copy and obtain any required permission.

Plan your comparison before submission

Mark one noisy phrase, one quiet phrase and a pause. After processing, compare those timestamps at similar voice level. Check complete words, natural tone and remaining distractions. A consistent room does not mean a model has been trained on that room, and no preset label guarantees recovery of a masked sentence.

Continue with the focused guide

This page remains at its existing address. For the detailed workflow, read the focused guide.

Before processing, keep the original, check your current plan and available minutes, and confirm permission to upload the material. Compare the actual words at similar playback levels, not just the size of the waveform. Download the accepted result before its displayed expiry and reopen the saved file.