Skip to main content
M
MHP777123
Inspiring
July 17, 2017
解決済み

Can I make one audio clip follow the peak levels of another clip across time (Q ported from Premier forum)

  • July 17, 2017
  • 返信数 1.
  • 754 ビュー

I am working on a classical music video that uses two audio sources. One was recorded using the venue's sound system using mics at the stage, uncompressed for peaks. The audio waveform therefore shows a beautiful sinuous pattern of softs, louds, and slopes between, and the performance sounds realistic, albeit realistic from an on-the-stage perspective. The other was recorded as part of the camera footage, taken from the back of the hall, to capture the ambience of the performance from the back-of-the-hall audience perspective; however, the camera automatically normalized all peaks, so that soft passages are louder than they should be and loud passages are compressed, resulting in an audio waveform that runs close to 100% normalization across time and sounds unrealistic. (I normally take my second audio source from a portable wav recorder, which also records uncompressed peaks, and mixing the two produces beautiful results, but I was unable to do that this time.)

I have edited both clips to be exactly the same length and have aligned them so that they start and stop at precisely the same points. The only difference therefore in the clips is that one has realistic peaks and valleys while the other has compressed and normalized these away.

So, I need a way to capture the peak PATTERN (over time) of the first recording and apply it to the second recording so that its waveform changes to LOOK LIKE the waveform of the first recording in its general sinuous shape. (The matching would be the general shape over time; one would not want to match frequency volumes millisecond by millisecond, because that would be unrealistic, since for example a timpani hit will sound louder and sharper on stage than at the back of the hall.)

Using Match Volume in Audition doesn't cut it, because I only get an across-the-board matching that applies to the whole clip. I cannot find any other sources discussing this, but it seems that with today's computing power and innovations there should be a way to do this. ???

このトピックへの返信は締め切られました。
解決に役立った回答 SteveG_AudioMasters_

MHP777123  wrote

If there aren't any programs out there that can do this, then I request it as a new feature! :-)

A feature like this has been a request for several decades now; it hasn't happened and I don't think it will. It has been found as a rule  that no realistic matching can actually be achieved, and the reason for this isn't the long-term sound level changes (where the envelope is easy to follow), but the rapid transitions. The effect of this is considerably worse than just listening to the original, as no algorithm exists that can apply a major dynamic change accurately enough. A lot of compression uses a square-law characteristic, and even if you can replicate the reciprocal law accurately, the results aren't good anyway - just like the old square law noise reduction units didn't work very well - exactly the same reason, and that is the unavoidable pumping effects that happen to the noise floor.

The most successful attempts to fix this so far have not used a reference file at all, but simply attempt to alter the gain structure of the file you're trying to repair to keep the noise floor constant. Unfortunately even this doesn't work very well either...

返信数 1

ryclark
ryclark
Participating Frequently
July 17, 2017

Unfortunately there is no way to do this since once the audio has been through the ALC (automatic level control) of the video recorder you can't recover the full dynamics of the squashed audio even if you could exactly match the inverse characteristics of the ALC's electronic circuitry. Obviously your close mic stage audio sounds too close. So the easiest way is to improve upon it is to add some suitable artificial reverb to it to simulate the effect of the hall acoustic.

M
MHP777123作成者
Inspiring
July 17, 2017

Yes, I agree about the reverb, it could give some simulated ambience. Even so, it doesn't quite give me what I want--I usually do the reverse. I have discovered that (for videos in Premiere Pro) pairing a stage recording with an ambience recording, and dropping the overall stage recording level down a bit, gives the ambience recording a much fuller audience-perspective sound without feeling like you're sitting on stage. (I normally capture both recordings, but I had technical problems with the ambience recording in this instance.)

Too bad there isn't at a way to do what I suggested: have an audio program create a general peak shape that follows the waveform and then apply the same shape to a file of identical length. Sure, the second source will still retain its compressed (i.e., super-focused) sound qualities in the parts that are very compressed, but for anyone who wants to do it this way (like me), it would save hours of time. The manual method requires comparing the two files manually section by section (using the visual dB grid to estimate average levels in each section) and then applying Keyframes in the second source (in Premiere Pro) to modify the Level of the second source over time, in an attempt to manually draw the shape I'm looking for.

If there aren't any programs out there that can do this, then I request it as a new feature! :-)

SteveG_AudioMasters_
Community Expert
SteveG_AudioMasters_Community Expert解決！
Community Expert
July 18, 2017

MHP777123  wrote

If there aren't any programs out there that can do this, then I request it as a new feature! :-)

A feature like this has been a request for several decades now; it hasn't happened and I don't think it will. It has been found as a rule  that no realistic matching can actually be achieved, and the reason for this isn't the long-term sound level changes (where the envelope is easy to follow), but the rapid transitions. The effect of this is considerably worse than just listening to the original, as no algorithm exists that can apply a major dynamic change accurately enough. A lot of compression uses a square-law characteristic, and even if you can replicate the reciprocal law accurately, the results aren't good anyway - just like the old square law noise reduction units didn't work very well - exactly the same reason, and that is the unavoidable pumping effects that happen to the noise floor.

The most successful attempts to fix this so far have not used a reference file at all, but simply attempt to alter the gain structure of the file you're trying to repair to keep the noise floor constant. Unfortunately even this doesn't work very well either...

契約条件Accessibility statement
Using the community Terms of use Privacy Cookie preferences Do not sell or share my personal information ad choicesAdChoices