How Descript Handles Mic Bleed in Multitrack Recordings
And what to do if it’s causing issues
Descript’s Multitrack Transcription, new in Descript 1.3, is a big step forward for anyone who works with speech in a multitrack setting.
The core principle behind multitrack recordings is to use a different microphone for each speaker. By isolating each speaker’s audio track, it becomes possible to make edits that wouldn’t be otherwise, like tweaking the volume of just one person. Multitrack recordings also facilitate Descript’s precise Automatic Speaker Labels — because the software can discern who said what, when.
But isolating audio tracks can be tricky. If you have multiple microphones in the same room, you may run into an issue that’s plagued recording engineers for decades: mic bleed (also known as mic spill). Mic bleed is what happens when your guest’s microphone can still faintly “hear” you — even though you’re speaking into a different mic across the room.
Note: mic bleed shouldn’t be an issue for Multitrack Recordings created from Zoom and Skype calls, because each speaker is in a different room — or a different continent. Just be sure participants use headphones or Bluetooth headsets to minimize feedback.
The good news is that Descript anticipates mic bleed and corrects accordingly. When you upload your multitrack audio files for transcription, Descript will automatically identify regions of duplicated text — which indicate that the same speech has been captured by multiple microphones — and removes them from the combined transcript (the text will still appear in each track’s individual transcript).
That said, if the mic bleed is sufficiently loud, it can lead to problems with your combined transcript, like words being attributed to the wrong speaker. If this is happening to you, there are ways to remove the mic bleed in your tracks before transcribing them.
To do this, you’ll need one of the many powerful Digital Audio Workstations (DAWs) available. The free tool Audacity is one option — as are popular DAWs like Logic and Pro Tools. In each case, your ally in this task is the Strip Silence feature (called Truncate Silence in Audacity).
Strip Silence will detect regions in an audio track that are quieter than the others (i.e., the periods when a speaker isn’t talking). These regions aren’t absolutely silent — they still contain room noise and the sounds of people speaking elsewhere in the room (which is what we’re looking to remove). After detecting these regions, Strip Silence will delete them, eliminating the mic bleed.
For more information on how to use Strip Silence in several popular applications, here are some helpful links:
Apple Logic X
- Apple — Remove silent passages in the Strip Silence Window
- Apple—Uses for Strip Silence
- You’ll probably want to adjust the parameters used by Strip Silence. The settings developer Marco Arment provides at the bottom of this blog post are a good place to start.
Pro Tools
Audacity
GarageBand
- Remove Noise and Hiss from your Podcast in GarageBand (This technique uses a noise gate, which is similar to Strip Silence)
And, of course, you can always take a more manual approach in a DAW by selecting and deleting the regions where mic bleed is visible in the waveform.
Once you’ve handled the mic bleed, remember to export each speaker’s track separately from your DAW before importing them into Descript; that way you’ll be able to take advantage of Multitrack Transcription with Automatic Speaker Labels (read this to learn how to create a Multitrack Transcription with Descript).
Here are instructions on how to export your tracks separately for each DAW.
Apple Logic X
Pro Tools
Audacity
- Export multiple tracks (you probably want to split files based on tracks, not labels)
GarageBand
