Fledgling voice artist seeking counsel.

[Paul L]

I think this is the right magic to calculate rms of the selection using the Nyquist prompt (the last sample of the sound is left out by integrate, but that is negligible), but I get a discrepancy of a fraction of a dB from Contrast.

I hope the bug is with me and not in Contrast.

Code: Select all

(let ((time (- (get-duration 1) (/ (snd-srate s)))))
  (linear-to-db (sqrt (/ (snd-sref (integrate (prod s s)) time) time))))

[steve]

Assuming that the track is mono and the selection is not too long, you can get the RMS value with one line:

Code: Select all

(linear-to-db (snd-fetch (rms s (/ (get-duration 1)))))

The maximum length that this works with is just over 2 million samples.

[Paul L]

I was avoiding the 2 million sample limit. My answers seem to be slightly more than from Contrast, and differences seem larger with time, but still tenths of dB or less out to 30 minutes.

[Paul L]

I too use Blue Snowball with adequate results. I am not expert enough to know what I am missing in a costlier mic. I just stand it on a card table. I use the cardioid setting with the back of it pointed toward the sometimes noisy street. I record without headphones so I know when airplanes approach, or vehices, and I wait out the interruption. I made a simple enclosure of hanging blankets with a roof and put the laptop outside. It sits on a cooling stand that draws power from the USB port, and those fans run quieter than the computer's own, and they run constantly, not intermittently, so that what they contribute to background is consistent. I use wireless mouse and keyboard and a monitor. I find an input volume of .4 right. My mic distance is a bit close and I use a pop filter.

All to show that I learned to mitigate noise at the source first, but low cost expedients can work.

[kozikowski]

I am not expert enough to know what I am missing in a costlier mic.

Maybe not much. I've used the one from work and it's a respectable performer. I use my Expensive Microphone® because it has multiple patterns and other controls and sounds very slightly crisper. But if someone said I had to shoot something with a Snowball, I would not be horrified. I would start looking for a graceful environment and bring my furniture pads with me.

Close is good. I think a lot of volume and noise issues could be solved by somebody presenting who isn't terrified of the microphone. On the other hand, it could be said that one reason for a pop filter's existence is to keep the presenter from swallowing the microphone. There is a sweet spot.

I think people are lulled into a false sense of competence by cellphones and conferencing aided significantly by not being able to hear themselves at the other end.

Koz

[Paul L]

As for voice distortion effects... I have used them, but I say, only if you are trying to be comical, or doing sci fi or fantasy. Even so it is a bit of spice that you do not want to overuse tiresomely. Maybe a bit of telephone like eq in serious stuff, but that is it. Distinguish males and females just enough to contrast -- a slightly higher pitch, a bit of breathiness. You don't have to be a perfect female to come across as a good dramatizer of all parts in a fiction. Listen to some audiobook greats like George Guidall.

Now this is artistic opinion, in which I am even less schooled than technical matters, so take or leave it as you like. But ultimately it is about pleasing ordinary people who listen, so I will speak as one of those.

[steve]

I was avoiding the 2 million sample limit. My answers seem to be slightly more than from Contrast, and differences seem larger with time, but still tenths of dB or less out to 30 minutes.

INTEGRATE loses a bit of accuracy as (apparently) it is calculated with single precision.
An alternative approach, less elegant than your solution but also less heavy on ram:

Code: Select all

(let* ((s (if (> len 2000000)
              (let ((step (round (/ len 1000000))))
                (snd-avg (mult s s) step step op-average))
              (mult s s)))
       (slen (snd-length s ny:all))
       (s (snd-avg s slen slen op-average)))
  (linear-to-db (sqrt (snd-fetch s))))

[Paul L]

Nice try. Pop and click removal is used to remove cat hairs from vinyl records. Those pops are usually distinctive enough to allow software to find them. Not so mouth noises. Those "look like" the rest of your show and voice and don't have a significant signature.

True, damage to the signal that looks like one big spike, is not quite like natural but undesirable click noises that do oscillate and decay on a fine scale, and the best math for finding and fixing each is perhaps not the same.

But I must disagree. There is a signature for mouth clicks, even if they are riding on another sound with bigger amplitude. They are obvious to the eye as well as ear, if you look at your sound in spectrogram. Clicks appear as bright vertical streaks. I figured out some math to automate the identification and filtering out of "bright vertical streaks" with some success. I intend very soon to post an updated version of my click removal tool, and as a bonus, a de-esser that almost fell out as an adaptation of much of the same code.

Ian, I recommend you get familiar with spectrogram as well as waveform views. They complement each other. Spectrogram helps you find where a click is more readily while zoomed out, but waveform helps you zoom in more precisely. I also prefer waveform dB view to waveform, because it stretches out the low levels and compresses the high levels, making little noises more obvious to the eye.

To use two views at once, try this: duplicate the track, mute the bottom one, make the top waveform dB and the bottom spectrogram. Click the pocketwatch ("Synch lock") so that deletion from one causes corresponding deletion in the other. Treat the top track as your master, the bottom as just an aid that you throw away later. Learn what various speech sounds look like in each view.

I'm remembering we got someone else past mouth noises a while back. It wasn't pretty. I think an elf designed custom software for his voice.

I first heard that term "Nyquist elf" from Koz, I think. Gee, it seems I have become one.

[MichloIW]

Greetings,

thank you for the further feedback.

I shall do some experimenting.

And no, I shan't be using distortion effects except perhaps for ONE thing in our current book and that is the few times our hero hears from someone via the radio but the signal is usually fairly bad. I'll see how that sounds but otherwise shall be sticking to what I can do with my own voice.

Cheers.

[Paul L]

I think that would not be overdoing it. There are some ready made eq curves to simulate radio or walkie talkie. Also try white noise from the Generate menu in a second track and mix it.