Before I render a lot of video, advice on the audio settings (dialogue), please

[lingyai]

I'm about to render many, many (300+) short videos for an online course I'll be launching soon. (About wind energy finance).

The audio, all dialogue, is now at the stage where I can't undo any DSP-ing done so far, but where I can -- because my video editor, which now has the pre-rendered audio stems, handles VSTs (but to a very limited extent as it's only so stable; also cpu-heavy ones make the video rendering way slow) -- have one last chance to make any finishing fixes.

Here is a sample, a boring 90 seconds grabbed more or less at random.

https://www.youtube.com/watch?v=b_6Kru5mBkI

My questions are:

Does it need any fixes? Things I've worked hard to deal with have been sibilance and noise reduction (not recorded in a brilliant room), as well as anything which might otherwise induce ear fatigue, while keeping compression gentle.

If it needs any, ahem, could they be fairly quick and easy moves, do-able with generic FX types? It's not that I'm lazy; were this a music track, I'd invest loads of time sculpting if need be; whereas, due to the way the work has flowed, to implement any last fixes now, I'd probably have to manually load any final fx chains manually, perhaps into 40 project files, so less really is more here. Say, a single EQ or something.

I'd be grateful for any advice. I'd ask that you listen to it on whatever device you'd naturally use when doing an online course of this type (need to see an Excel screen, so probably not a phone), as opposed to your best audiophile cans etc.

[James Perrett]

To be honest, to my ears it sounds too processed already and I think you've gone overboard on the de-essing. I'd be interested to hear the raw audio for that clip.

[RichardT]

Yes, I agree, the S sounds are a little slurred, and it’s starting to get a little quacky. If you need to live with what you’ve got, I’d go with it as is.

[lingyai]

Thanks gents.

Is this an improvement?

https://www.dropbox.com/s/2jicna50znj0r ... 3.wav?dl=0

[James Perrett]

That may be very slightly better but still over processed to my ears.

[resistorman]

To be blunt, I couldn't take listening to this for more than a short period of time. It's over processed so far it's uncomfortable.

[Mike Stranks]

I've not listened myself, but have you come across the videos/livestreams/courses of Curtis Judd?

His specialist subject is audio for video. he and his associates really know their stuff and he talks and presents in a friendly, relaxed way with no hype and no histrionics.

It sounds like a restart on what you have may not be possible, but should you need to go down a similar route in the future...

Also, iZotope have lots of very helpful stuff on their website including the order that you should apply remedial processing.

Good luck!

[Tim Gillett]

I'd be interested to hear the raw audio for that clip.

I would also like to hear the voice before processing.

At times the the denoising/gating clips off the ends of words but there's no hint of any room reverb, suggesting the denoising could be backed off somewhat. Over denoising is listed on the Izotope website as one of the most common mistakes in audio processing. As Mike said, Izotope has some great tutorials on using their products, including before and after examples. https://www.izotope.com/en/learn/rx.html

A spectral display reveals a number of deep comb filter like valleys, probably making the voice sound less natural. Again, just an unprocessed voice sample would give a better idea as to what is happening here.

[lingyai]

If I were able (time-wise) to start afresh, the raw audio would indeed be a useful starting point. Whereas here, like Hedwig, I'm doing my best with what I have left to work with.

Not quite clever enough to be fully daunted, here is another go.

https://www.dropbox.com/s/n8ck10gis6rjm ... 5.wav?dl=0

Used TDR's Nova dynamic EQ to work on the sibilance. For me the sound is now darker but to me at least intelligible, and ok for an instructional video, where the aim of the voice is to explain but do no harm, i.e. not be off putting.

BTW, please don't judge by the short phrase where I lisped (cottonmouth), saying "even though each of the stepth themselves are simple" as that is atypical, and not from processing.

[James Perrett]

I think the lesson here should be to keep all the intermediate stages of audio processing (or at least keep the original audio and all the settings used in the processing). Also don't process your own voice. Many people seem to hate the sound of their own recorded voice so they will try to make it into something that it isn't and end up over processing.

The new one is slightly better again but still suffering from too much de-essing.

[Kwackman]

the aim of the voice is to explain but do no harm, i.e. not be off putting

I hate to say this, IMHO that processed voice IS off-putting. It almost sounds like an older generation of AI voice.
Other opinions may be available.

[Mike Stranks]

Hmmm.... difficult answers time...

Depending on the profile/'importance'/earning potential/etc of the course then it might be prudent to start again. I know you don't want to hear that, but it seems like you're now trying to rescue something that's beyond rescuing... a bitter pill I know.

But if you are going to start over, come back here before you start. There are several here whose 'bread and butter' is spoken word - with or without video. Thus you can gain useful insights about getting the source right so that the amount of post-processing required is reduced as much as possible.

I know you don't want to hear that, but sometimes starting over really is the best solution...

[lingyai]

Maybe too subjective a thing to have expected a quick-use consensus steer on ... FWIW I've asked here, a trainer in the same medium, and 7 prior students, i.e. folks who by definition are my target audience. The only ones who said it is a barrier are here. Plus there's my own ears, which after these iterations don't object. So not sure I'd declare it a smoking wreck, but thanks for the views.

[BWC]

How long did your other test subjects listen for? You expressed concern about it being fatiguing, and I'd agree with the others here that it is. I'd suggest having your target audience test subjects listen for as long as you expect the actual audience to listen for, and see if it gives them any trouble. Not having the level of ear training that folks around here tend to have, they might not notice a problem so immediately, but only after the fatigue sets in.

[resistorman]

the aim of the voice is to explain but do no harm, i.e. not be off putting

I hate to say this, IMHO that processed voice IS off-putting. It almost sounds like an older generation of AI voice.

My thought exactly. Extremely grating.

[resistorman]

Maybe too subjective a thing to have expected a quick-use consensus

I think the consensus is quick and clear. You have made your narration unlistenable. Learn from your mistakes.

[lingyai]

Well, I guess "the consensus is clear" if you exclude over half the respondents.

The argument that those who are fine with it actually don't know enough to trust their ears, is not so persuasive for me. It's like saying even if someone says they like a sample of chocolate, they are wrong because they are not educated/experienced enough to know what they actually like. A minority of us use DSPs, but everyone watches videos with audio.

I didn't ask people how many times they listened, I trusted them to decide for themselves, just as I've done here.

I've posted 3 versions here, responses would be more useful if folks said which they were responding to.

Anyway, the discussion did prompt me to try some other things. This 4th version is where it stands now. It's 12 seconds of speech, played twice in a row.

https://www.dropbox.com/s/29iaqqdstryw9 ... 3.wav?dl=0

(The delivery is a bit halting as it is syncing with some screen action)

I've listened on several devices, higher and lower end, slept on it, and then listened again.

I for one truly don't hear, for example, anything resembling an early stage AI voice, as someone put it, but horses for courses.

To my ears this version tames a lot of the original sibilance while improving articulation vs the previous versions, and sounds pretty much like normal if boring speech of the sort one hears in many instructional videos, which I've consumed a lot of without auditory objections.

So in getting me here, the discussion has been a useful prompt, thanks.

[RichardT]

I think the likely cause of what’s happened is that folks on the SOS forum spend a lot of time listening to and fixing up audio so they are going to be more critical than the average listener.

[Mike Stranks]

Although I trust the comments of some of the previous posters here - I know their background and how they earn a crust - I decided to listen for myself to determine my personal views: are people being hyper-picky and is it fuss about nothing or are there fundamental issues?

It's the latter.

I've been working with spoken word for several decades and immediately heard several issues.* And the OP suspected there were issues which is why he came here in the first place.

But he is where he is. He doesn't want to start over, and states that his target audience is quite happy with the sound. So end of discussion.

But is it good spoken word? No.

* In summary: the results of over-processing. Been there, done that, got the T-shirt.

[Kwackman]

I for one truly don't hear, for example, anything resembling an early stage AI voice, as someone put it, but horses for courses.

That would be me, although I did say there would be other opinions...
I was commenting on the latest version that was available when I posted.
If you're happy, then go with it.

There's another thing that might be worth considering.
We're listening to WAV files here.
If this gets streamed, the audio will be processed to the streaming platform's format. So, that might change the quality of the audio.
This is not an area I'm knowledgeable about, hopefully someone who is can confirm/advise.

Apart from the subjective audio quality, I like your "performance" (I'm sure there's a better term for it!). You're not shouty or over animated which is a pleasant change from many video voice overs out there.

Good luck with your project.

[BWC]

The argument that those who are fine with it actually don't know enough to trust their ears, is not so persuasive for me. It's like saying even if someone says they like a sample of chocolate, they are wrong because they are not educated/experienced enough to know what they actually like. A minority of us use DSPs, but everyone watches videos with audio.

An average person might need to take a couple of bites before noticing a funny aftertaste, whereas someone who taste-tests chocolate all day, everyday, might only need a quick smell to know that something's off. This is not meant as an insult to the average person's taste buds; it's the reason why you ask professionals for their advice. You are, of course, free to ignore that advice, at your own risk.

[Drew Stephenson]

For what it's worth, listening to the last version posted, I do think there is still too much de-essing going on, but when it's been compressed for a video file, throttled by network management tools, and played back through a set of laptop speakers or cheap earbuds, it'll be fine.
I'd run with it and chalk it up to experience for next time.

[Tomás Mulcahy]

Consider that the people you asked first are personally connected with you so may not be entirely objective. There was a reason you then decided to post the question here. What was that reason? Did you suspect, at some level, a lack of objectivity?

Whereas the users here are being objective because they are responding purely to the question. Not to you personally. They have no skin in the game.

Consider also that when you've put in a lot of work you won't want to (seemingly) throw it all away. Especially with a deadline looming. I think that's called disappearing down a rabbit hole? Happens to us all! Personally I take these occasions not as wasted time, but time spent learning a valuable lesson.

[Drew Stephenson]

One other thing I have just remembered. I think it was a Harvard study that said a subject matter expert could seem up to 20% less credible if the audio on a lecture was bad. I wish I could find the paper again.