Podcasts, interviews and meeting recordings · Updated 1 October 2026
Reduce the noise.
Hear the conversation.
A cleaner recording is only useful if the speaker still sounds like themselves. Compare the voice as carefully as the background: quiet syllables, breaths and the ends of words can matter more than total silence.
Upload an audio file or supported video from your device. This tool cleans a saved recording; it does not connect to a live meeting.
Choose the tool for the unwanted sound
Hiss, hum or recording noise: start with Noise remover and compare its Cleaned and Original sound.
A music bed or a song in the room: compare Voice isolator. Music separation and noise cleanup are different jobs.
Echo, overlapping speakers or dropouts: there is no guaranteed repair here. Separate participant tracks, when you have them, give you a better way to work on each speaker. Missing audio cannot be recovered just by reducing noise.
Listen: speech with added hiss
The before is a cleared speech recording with synthetic hiss added. The after is actual output from the noise-cleanup engine used by Mirinex. Both use the same gain, without an after-only volume boost.
This demonstration uses telephone-quality speech with added hiss, not a real podcast or Zoom call. It cannot show how a studio recording would sound. Source and changes.
A cleanup workflow for your actual recording
- Keep the original. Start with the highest-quality saved recording available. If you have separate participant recordings, work on the noisy one first rather than altering every speaker.
- Pick a representative passage. Include ordinary speech, a quiet word and a pause where the noise is audible.
- Choose the file in Noise remover. The limit is 20 minutes and 250 MiB per file. For an hour-long episode, export shorter sections from your editor. The audio cutter can prepare an audio excerpt; it exports audio, not a trimmed video.
- Switch between Cleaned and Original. Check that consonants, breaths and soft syllables survive. If the result sounds too stripped back, raise Removed sound a little to blend back part of the original.
- Download and listen in context. Audio downloads as WAV. Supported video can retain its picture with the new mix, with WAV available if video export is unsupported. Check joins with untreated sections before finishing the episode.
What counts as a better result?
Listen at a comfortable, similar level. Is the background less distracting? Are words complete? Does the person still sound natural? A silent pause is not a win if the next word sounds watery or has lost its beginning.
A recording that has already been heavily cleaned may not improve with another pass. Keep whichever version works better in the finished conversation, even if that is the original. If noise removal struggles, you can compare voice isolation on the same source.
If the voice is clean but too quiet, loudness normalization may be the next step. It changes level; it does not remove background noise.
Hear your recording cleaned up
Choose the episode, play the difficult passage and compare it with the original. Each file can be up to 20 minutes.
Hear the first minute of your own file. A Day Pass or Pro processes the whole thing.
When the problem is music
See music removal for a video upload or a speech recorded over wedding music for the separate voice-isolation workflow.
Demo source and changes
Speech: Open Speech Repository, American English female recording OSR_us_000_0010_8k.wav. The repository permits modification and before/after technology demonstrations with source credit.
We used the first 12 seconds of the 8 kHz source, resampled it, added synthetic hiss and processed it with MossFormer2 SE 48K. Both samples are 44.1 kHz stereo WAVs at identical gain. Resampling does not add missing high-frequency detail. No separate enhancement, after-only normalization or extra soundtrack was applied.