How Webex Fixed My Cacophonous Interview

by Dave Michels

Videos are more complex to make than they appear. I’ve been writing about enterprise comms for close to two decades. It’s an iterative, asynchronous process. The backspace key is the most used key on my keyboard.

Producing on-site videos is complex. It takes a lot of effort to get it right – either during the filming or during the editing. I think that’s one of the attractions to livestreaming – there is no editing (and it’s a feature).

Last week I attended the Five9 analyst summit in Porto, Portugal. It was a “well-orchestrated” event. Analyst events are among my favorite discussion topics. There are good ones, bad ones, and a new emerging category of effective ones. Post-pandemic analyst events are shorter, so they have to be more effective. Between two dinners, Five9 hosted an extremely effective event. Effective for them, for analysts, and I expect in terms of their investment.

There was really only one big snafu at the event, and it was mine. It was the last evening of the last event when I tapped CEO Mike Burkland to make a video. He agreed to the impromptu shoot, and within 10 minutes I had my camera set up and my lapel mic on (mistake 1). Dinner was loud (even had music), so we used an adjacent room that seemed quiet (mistake 2).

it was late and we were all tired, so I decided to narrow the focus of the interview to vision and mission, which was part of his earlier presentation. As soon as we got started, people began flowing into our space. Let’s call them Five9 revelers full of wine and joy. At one point, I was sure that a tour bus was unloading. No fear though – I was after all wearing a lapel mic.

It turns out wearing a lapel mic isn’t enough. You have to properly pair it with the recording equipment first. The built-in mics stepped up to the job, so audio was recorded, but it wasn’t honed in our voices. There was a lot of noise in this recording. I posted it anyway, and immediately got crap from my peers. One said “why don’t you invest in a mic?” “I did,” I replied, “you can see it on my shirt.”

My original recording actually had multiple sound problems.

  1. There was a lot of noise from passer-bys.
  2. Because the mic was far away, there was additional reverberation that muddled clarity.
  3. There was a lot of variation in sound level. Mike and I occasionally faced the camera while speaking, so we went from loud to faint with a turn of a head.

This is what professionals like me call an audio mic malfunction (pardon the jargon). The question on my mind now is if the video can be saved at all. 

I tried to fix the audio by running it through some ANC software. ANC, as in Active Noise Cancellation, is actually the wrong tech. ANC uses multiple mics to separate source from noise and dynamically create a sound filter to block the noise. It works well on headphones. ANC is irrelevant to this use case as my noise and signal are combined on a single source. I also tried some other tools associated with noise reduction utilities that were built into my editing app to no avail. 

With everything now on one track, I needed a way to detect and remove the noise. Something I remember Cisco talking about with its 2020 acquisition of Babblelabs. The problem is Cisco Webex isn’t designed for file uploads, though I have since learned Webex Vidcast supports uploads. I presented the video as a Webex meeting participant.

Meeting solutions typically don’t use multi-mic ANC solutions such simply because most users only have one mic. Instead, they leverage emerging machine-learning techniques to manage noise. High-end conference rooms do use multiple mics, but there the tech is optimized around speaker tracking instead of noise reduction. That is multiple mics in conference rooms separate speakers, not speaker(s) from noise.

The Webex/Babblelabs tech was amazing and removed most of the noise. As advertised, it was magically able to identify the noise, separate it from the signal, and remove it. It also improved the reverberation issues. Being able to remove noise is reasonably recent technology. Noise reduction techniques, such as those associated with Dolby, reduce noise but don’t remove it.  

Check out this sample reel:

For grins, I tried the same experiment with one other competitive meeting solution (to be unnamed). Not only was it not as effective at noise removal, but when I ran its output into Webex as an input, Webex was much less effective. Although it did reduce some noise, it also introduced artifacts that Webex could not repair. The best result was the naturally noisy recording run through Webex. 

Cisco has integrated the Babblelabs technology across the Webex portfolio. This includes meetings, contact centers, and even PSTN calling. Most importantly, it’s also integrated into most of the Webex endpoints, including Webex Desks devices. The Webex Desk and Webex Desk Pro can function as a USB camera, mic, and speaker. As noise reduction is built into the hardware, users can get Webex noise reduction on almost any meeting system.

It made me realize that almost every major meetings solution advertises some form of noise reduction, but they are not equal – and difficult to compare without trying them. There are some industry terms like PESQ scores, but these scores don’t necessarily align with human-scored results. 

It also makes me scratch my head that these technologies are locked into meetings applications. For example, I typically use several tools for transcription and noise reduction that are not as effective as Webex, but are designed for upload files. It would be nice if Cisco would make these tools more readily accessible as part of the Webex Suite. 

The entire interview with Five9 CEO Mike Burkland is here. That embedded video also contains a link to the original, noisy recording. 

Hat tip #1 to David Maldow for his expertise in workarounds.

hat tip #2 Cisco Webex Team for guidance and explanations.

Hat tip #3 to the fictional Dr. Kakofonus A. (for As-Loud-As-Possible) Dischord. He is a doctor of dissonance that lives somewhere in the forest near the Valley of Sound in a carnival wagon. He has an assistant named The Awful Dynne.