Last week, when I was explaining how I so badly borked the audio file for the previous week’s show, I profusely thanked Michael Westbay and Aris Skamagakis for telling me it was messed up. I said that I always want you to tell me when there’s a problem with the audio or podfeet.com. Sometimes I can’t fix things, but if I can, I’d sure rather know!
Inspired by that encouragement, Donald K wrote in this week with something I never knew wasn’t working as-designed. He tells his story of discovery of the shows first, and it’s a fun story so we’ll let him start at the beginning and then you’ll hear him explain the problem he found
Donald K wrote:
Hey, Allison –
I discovered your shows while searching for discussions about JavaScript and, of course, found you and Bart through the unique and helpful Programming By Stealth. Across all of your shows, I really appreciate the time and attention you devote to discussing systems and process.
One thing that prompted me to write today is something I’ve been curious about since discovering NosillaCast. It took me a few episodes to realize what I was hearing—your cash register “cha-ching” sound effect. It’s been consistently very quiet, nearly indistinguishable. At first, from what I heard, I imagined it was being triggered and played live, just barely bleeding through a directional, background-suppressed microphone without a direct signal to the recorder. I doubt this is the setup, but is how I’d describe it.
Since I assume this isn’t fully intentional—my first guess being that you’d want it to match the volume of the other transitional jingles—and in light of your recent call for listeners to speak up if something doesn’t seem quite right, I decided to say something on the off chance you weren’t aware. Either way the show is great!
Have a good day!
-Donald
I love everything about what Donald wrote, especially that he found us through a search for JavaScript instruction which lead to Programming By Stealth which lead him to the NosillaCast.
Now about that “cha-ching” sound effect. To my complete surprise, Donald is 100% correct. If you’ve got the hearing of a Doberman, you might be able to hear the “cha-ching” sound I’ve been putting before I panhandle for financial contributions. I paid a whole $4 for that jingle and had no idea that it was nearly inaudible.
When I create the show using Hindenburg, I drag that jingle into a track separate from my voice track, and listen to it, and the volume is definitely at the same level as my recordings. I went back to the exported AAC file for the last couple of shows, and it sounds about the right level there, too. However, when it goes through Auphonic’s process, it brings the audio way down.
My first theory was that the cha-ching sound is so short that maybe Auphonic thinks it’s noise, so it brings the level way down. That got me worried that maybe you haven’t been hearing Bart’s audio icon I play after Security Bits, either. It’s about the same length. I listened back to a recent episode, and luckily, that lovely tone was audible as expected.
I was still thinking it was too short, so I went hunting for a new jingle that was a bit longer, but was still the sound of a cash register. Maybe that’s a very old-school sound, but it’s the only one I can think of to make you think of money. I like Audio Jungle because you can buy individual sounds and songs for not a lot of money. I think Jill from the Northwoods recommended it to me. Anywho, I was having trouble narrowing down the options Audio Jungle offered to me, so I decided to enlist Perplexity to help me find what I wanted.
My prompt to Perplexity was:
I’m looking for a sound effect to play before the part in my podcast where I ask for financial donations. I’d like something that evokes the thought of money. I want it to be around 3 seconds long. I have an account at Audio Jungle where I’d prefer to buy it.
I’m glad I was so verbose in my question, because after providing me with a few links to Audio Jungle with jingles that met my criteria, it also offered some integration tips for my podcast. It suggested keeping the sound effect at or below my voice peak (which isn’t the problem I have), but it also said to consider a tiny 100–200 ms pause after the sound to let it “land” before I start talking.
I went back to the original recording in Hindenburg and took a look at when I started talking after the jingle. I clearly start talking well before the jingle completes. This led me to believe that perhaps Auphonic is ducking the audio because I start talking before it finishes.
It was time to run an experiment. I created a new audio file in Hindenburg, recorded a little intro, and then did the following tests:
- Short money jingle where I started talking before it finished. This should replicate what you haven’t been able to hear during the show.
- Short money jingle where I waited till it’s completely finished to start talking. If the jingle is audible after processing through Auphonic, that would mean it’s the ducking problem. If it’s not audible, it would mean the problem is that Auphonic thinks it’s noise.
- Long money jingle where I start talking before it finishes.
- Long money jingle where I wait till it’s completely finished to start talking.
And the results are in. After the file went through Auphonic, you can’t hear any of the jingles in the four tests! This confirms that Auphonic must see this type of sound effect as noise.
I looked at the settings I’ve chosen for my presets in Auphonic. There’s a section entitled Audio Algorithms. Under Noise reduction, for the denoising method, I have it set to “Dynamic: keep speech and music; remove everything else”, with a noise removal of 30 dB. Since this “cha-ching” sound is clearly not speech and clearly not music, I think I’ve found the problem.
The other Denoising Method options are:
- Classic: legacy denoising algorithm
- Speech Isolation: keep speech, remove everything else
- Static: remove constant noises only
Legacy doesn’t sound good, and removing everything but speech is even more heavy-handed. I could go with Static to remove constant noise only, but then if the garage door under me opens while I’m recording (which happens more often than I’d like), or someone bumps their mic (probably me), or if Kepler the wonder dog shakes his collar outside my door, those sounds would not be removed.
The other option would be to play around with that 30 dB noise removal level. I gave Perplexity the two MP3 files, before and after going through Auphonic, and asked it for a recommendation on the noise removal level that would keep the jingle. She suggested I test at 10dB and see if the rest of the ambient noise level is too high. Then she talked about messing around with options at 15 and 20dB. I asked her for a mathematical answer to stop removing this particular sound that wouldn’t mean a bunch of testing and guessing. She told me that a mathematical answer wasn’t possible without knowing exactly how the algorithms work at Auphonic.
Perplexity then asked me to describe my environment and whether noise was a problem or if I had really clean audio to begin with. I was about to write back to say that my audio is pretty clean and then remembered that we embed the audio from the interviews on very noisy show floors at CES and CSUN.
Now this is where this entire adventure becomes comical. Remember how, when I borked the file, it was because I was experimenting with various versions of interviews from CES with and without noise removal? I told you that I did this very serious A/B test with the live show audience to see which one sounded better.
But if Auphonic is doing noise removal so aggressively that it entirely removes my “cha-ching” jingles … what’s happening to the interviews from CES?
You probably already know the punchline to this story. I played a recent episode, and sure enough, there is ZERO show floor noise during the CES recordings! All that faffing about with noise removal during export from Final Cut Pro was 100% wasted energy, as was the A/B testing with the live audience.
I always check the audio file before spamming the Internets about the latest episode, but I don’t sample every chapter, so I never noticed this was happening. I said the show floor noise was completely gone, but technically, there’s kind of a “whoosh” of noise right before the interviews start, because the beginning is overlaid on the jingle before it starts. My hope is that if I let the jingle “land” as Perplexity suggested, maybe it won’t whoosh at you. But maybe Auphonic needs a second of sampling before it can remove the noise. We’ll find out!
Steve and I discussed whether we should change the threshold for noise removal by Auphonic so you’ll be able to tell I’m not in a sound booth but rather on a noisy show floor, and we decided we’d leave it as is.
As much as I love the “cha-ching” sound of a cash register to let you know the panhandling is about to begin, it was time to search for a new audio file that was music. Back to Perplexity to ask it to find me something musical that evokes the desire for people to send me money. After listening to a few, I narrowed it down to two that I liked and asked the Podfeet Slack community to vote. By a two-to-one margin, I bring you “Simple Soft Tone“. (You’ll hear the audio watermark “Audio Jungle” on top of it which won’t be in the show.)
Bottom Line
The bottom line is that this has turned into a comedy of errors on my part, but I never would have known about the jingles disappearing all this time if Donald K hadn’t told me! You know I love to dig into problems and understand the root cause, so this has been a fun adventure for me.
In the end, it cost me $5 for the new, longer cash register jingle I can’t use and $10 for “Simple Soft Tone” that you’ll hear before my panhandling segment! I’m losing ground on this whole making money deal.
Remember simpler times when I just paused between segments with no audio jingles at all?