It can be applied either before transcription to improve accuracy or after, depending on the workflow.
No. The original audio timing is preserved to maintain synchronization.
No. All denoising is performed locally without uploading audio data.