Auphonic vs Human Audio Mixing: What Each One Actually Fixes

For a lot of shows, yes, Auphonic on its own is good enough. If you record one or two people on decent microphones in quiet rooms and your only problem is that one voice is louder than the other, Auphonic will solve that for a few dollars an episode and you should just use it.

Here is what it is genuinely strong at. Loudness normalization to a target like -16 LUFS for stereo or -19 LUFS for mono, which is what most directories expect. Adaptive leveling across a whole track so the quiet guest comes up without the loud host clipping. Steady state noise removal, meaning fan hum, air conditioning, laptop whine, anything constant. Filler and silence cutting on the newer tiers. Multitrack processing with crosstalk gating, so the host mic stops picking up the guest through the room.

Auphonic solves loudness. It does not solve judgement. It cannot hear that your guest said something brilliant while a truck went past, and it cannot decide that the best move is to keep the line and duck the truck rather than gate the whole passage into a robotic wobble. That is the honest split. If your raw audio is clean, buy the credits. If your raw audio has real problems in it, an algorithm applied evenly across 58 minutes is going to make some of those problems louder.

Most hosts land somewhere in the middle, which is why our editors often run something like Auphonic as one step inside a bigger pass rather than treating it as the whole job. The wider version of that argument lives in human podcast editing vs AI tools.

A human fixes anything that requires knowing what the words mean. Auphonic processes a waveform. My editors watch the whole episode, so they are reacting to content, not amplitude.

The list is fairly consistent across the shows we cut. Room echo and slap, which is reflection rather than steady noise, so a broadband denoiser either ignores it or chews the voice trying. One-off bangs: a mug on the desk, a dog, a door, a chair creak under a punchline. Plosives on a mic with no pop filter, which need to be handled per hit with a low cut or a redrawn breath, not a global filter. A guest who drifts off axis for four minutes then comes back. Two people talking over each other where the fix is a choice about who wins, not a gate threshold.

Then there is the stuff that is not repair at all. Music beds that duck under the intro and lift on the outro at the right emotional beat. Bleeping a name the guest asked to pull after recording. Matching a phone-in caller to studio audio so the jump does not hurt. Keeping a cough in because the room laughed at it.

The honest test is whether your problems are constant or occasional. Constant problems are software problems. Occasional problems are human problems, because they need someone to notice each one and decide. Our editors work through the timeline in passes, which I break down step by step in the podcast workflow guide.

Send us the raw files and skip the processing. That is my real answer. But if you want to run Auphonic yourself first, or you are staying DIY on audio and only sending video, here is what I would set.

Upload multitrack, never a pre-mixed stereo bounce. One file per microphone. Once tracks are baked together nobody can separate them again, and a bad mix becomes permanent. Set adaptive leveler on, but pull the strength down rather than leaving it at default if you have a dynamic speaker who whispers for effect. Noise reduction on auto rather than a fixed heavy amount, because a fixed amount applied to a clean track adds artifacts for no gain. Loudness target -16 LUFS stereo, true peak -1 dBTP. Turn off filler word cutting if the episode is also going to video, because the audio and the picture will drift out of sync and someone has to fix that by hand later.

Export a lossless WAV, not an MP3. If you hand an editor a 128 kbps MP3 that has already been leveled and denoised, we are polishing a compressed file and every later move sounds slightly worse.

If you record on a browser platform, the quality of the source files matters more than any of this. I compared how two of the common ones handle local recording and backup in Zencastr vs Riverside.

Auphonic is cheap. Credits work out to a small per-hour cost and you can process a weekly show for the price of a coffee. There is no argument to be had there, and any comparison that pretends otherwise is selling you something.

Our pricing is per episode. Starter is $199 per episode: single camera, 1080p, one revision round. Studio is $299 per episode: two to four cameras, 4K, unlimited revisions. Both cover up to 60 minutes as the base. Up to 30 minutes costs less, up to 90 minutes costs more. A 4-episode pack saves 15% and monthly saves 25%, so a weekly show on monthly lands near $149 an episode on Starter.

That is not just audio. It is the full cut: audio repair, picture, camera switching, music, titles, exports. Compare it to a typical agency at $300 to $400 per episode plus roughly $400 a month once you add clips, and the gap widens rather than narrows.

Add-ons are flat too. Vertical clips are $25 each, show notes and chapters are $35, rush 24-hour delivery is $75. Everything is listed on the pricing section so you can build your own number before you talk to me.

So the fair framing is not Auphonic at $3 versus us at $199. It is Auphonic plus your own three hours versus us plus your zero hours. Price your own time honestly and the comparison gets a lot less lopsided.

Three places, and I will name them plainly.

Speed. Auphonic returns a leveled file in minutes. Our first cut comes back in 48 hours, with revisions the next working day, and that is fast for a human process but it is not minutes. If you record at 9pm and publish at 6am on a news show, software wins.

Consistency across a back catalogue. If you have 180 old episodes at wildly different loudness levels and you want them all normalized before you move host, batch processing is exactly the right tool. Paying a human per episode to re-master an archive is a bad use of your money and I will tell you that on the call.

Cost at volume. A daily show with clean audio and a tight budget should automate the mastering step and spend the saved money on something with more upside, like better microphones or short clips that actually bring new listeners in.

What I would not do is stack automation on automation. Auto-level, then auto-cut, then auto-clip, then auto-thumbnail, and by the end nobody has heard the episode before it went out. That is how shows publish a dead air gap or a half-sentence clip. The same trap shows up on the video side, which I wrote about in Opus Clip vs human clip editing.

No mystery here. The pass is roughly the same every week.

First, we listen to the whole episode with the multitrack open. Not scrubbing, listening. Every problem gets a marker. Second, we clean per track rather than per mix: gate the host mic where the guest bleeds through, low cut where a plosive lands, spot repair the bangs. Third, leveling, which is where a tool in the Auphonic family often does the heavy lifting because it is good at it and I am not going to do worse by hand out of pride. Fourth, the creative layer: music in and out, transitions, ducking, any bleeps requested. Fifth, a full listen top to bottom at conversational volume on normal headphones, because that is how your audience hears it.

We also match audio to picture. A cut that works in a waveform can look wrong on a face, and on multi-camera shows the switch has to land with the sound. That is a big part of why we handle both together on multi-cam editing rather than treating audio as a separate vendor.

Nothing is auto-cut by silence detection. Every cut is a decision someone made. Hosts tell us the difference shows up most in the pauses, because software removes the breath before a hard line and a human leaves it in.

One last thing on ownership. Footage, project files and exports are yours. If you leave, you leave with the sessions.

Yes, and that is what most of the shows we cut end up doing. Software for the mechanical work, humans for the decisions. The two are not really competitors, they are different layers of the same job, and the marketing on both sides muddies that.

A sane setup looks like this. Record multitrack locally at 48 kHz WAV. Run automated leveling to a fixed loudness target so every episode in your feed matches. Then have a person do repair, music, structure and the final listen. If you are DIY on video but drowning in audio cleanup, do the reverse: automate what you can and hire out the pass that needs ears.

The decision I would push back on is skipping the human listen entirely on an interview show. Interviews are where the value sits in the moments nobody planned, and those are exactly the moments an algorithm treats as errors.

If your show is one person, one mic, one room, you may not need us at all yet, and I would rather say that than sell you something. There is a plainer breakdown of that setup on our single camera page. And if the packaging is your real bottleneck rather than the sound, my sister studio ClickTheory does YouTube thumbnails at 8 for $250, which will move your numbers faster than another decibel of noise reduction.