Shooting with 2 different mics and background noise

[jimmyy]

Hi,

My first trial with shooting some scenes were far from perfect.
I recorded two persons on the street, one of them had a lavaliere wireless mic, and another did not. I had to rent all the equipment, and they did not have two wireless mics.

Now I have a movie with significant background noise, and I could extract the sound(via VLC conversion to flac format) and split it in two tracks in Audacity. One track is ok, the one on the wireless mic, but the other one captured a lot of the background noise.
It looks like the separation of the tracks resulted in two mono tracks.

I've used an Sony camera and two XLR mics ( a shotgun on the camera) and an wireless mic. I would have expected that each of the two tracks are stereo, but maybe this is how they were recorded, or maybe I do not know how to use audacity that well.

The background noice is for one shooting cars passing by, and in another one some background music.

How could I improve it and minimize to the maximum the background noise?

I've tried the noise reduction effect in Audacity, with some minimal improvement...

The difficulty is that I can't shoot it again, it costs quite a lot to rent the equipment, to find the people... to fly there... (I believe that you know better than I do). This was my first time.

Thank you for your advices

[Khaver]

If the two people are in a conversation try to clean up the noisiest track as much as possible. Make a loop track of as much of the cleaned up noise without the person talking and use this under the track from the lav mic so the two will sound similar. You can also try gating the noisy dialogue and the running a background track under both of the dialogue tracks.

You could also try replacing the dialogue with new voice talent.

[Cornucopia]

The problem with true, real-life NOISE is that it is random. This makes it particularly difficult to do something as simple as subtract out (mixing out-of-phase). Plus, even though it is officially called statistically "random", it has local variations that make it uneven to filter against.

Try uploading a 10-30sec clip of each of the 2 tracks in isolation (or if they are separate L+R of a dual track, send them that way and describe them). Please, DO NOT send them pre-mixed. We can try our hand at it, or give suggestions...

First, let me give you a few hints about NR:

1. ISOLATION is the name of the game. If you go too extreme, however, you will lose quality in the desired signal along with losing the undesired signal (noise). It is best to do this in MANY passes, if necessary, so that each pass does only minimal damage. Think of "chipping away at block of stone with a scalpel instead of a sledgehammer to get to the bust inside".
2. When you isolate, you must first analyze what all the elements are combining to give you the composite sound that you recorded. Analyze them in ENGINEERING terms: volume, frequency, dynamics, phase, formant, spectrum, delay/echo, etc. See how the signal that you desire compares with the signal you don't. Use those terms to find clearly where there is some wide-enough difference in the demarcation between the good and the bad signal - and then use tools that work on those features.

For example: voice has a limited spectrum (~80-6000Hz, less if speaking). If you clearly do not care about other signals beside the voice, a simple voice-optimized static bandpass filter will filter out other frequencies that fall outside of that bandpass. This then makes signals that might be partially in and partially out of that bandwidth be less in volume compared to the (relatively) untouched voice volume. This ratio change can then be taken advantage of in the next stage (like I said, multiple stages).

Note, that if you DON'T want to (or can't because of logistical reasons) do all passes in realtime in sequence (daisy-chaining), you should ALWAYS use uncompressed/losslessly-compressed files as intermediates.

Some suggestions for the future:
1. You get optimum audio signal quality (in terms of SNR) from these factors (in roughly descending order of importance): Mike Placment/Distance, Mike PolarPattern/Sensitivity, Subject Loudness, Environment, Mike internal & subsystem Quality, and Mike Physical Form-factor.

Note that placement is at the top. You can maybe get GOOD but you likely never get GREAT audio with a shotgun mike that is camera-mounted, because that is NOT the optimal place for the microphone. That's only an optiimal place for the videographer. This is why BOOM poles are so important, becaues you can manually & instantaneously move the mike into the best position possible (within the usual parameters of not being visually noticeable, etc). Lavs are good also, but their form factor often limits their quality & sensitivity. And their placement is very tenuous, having good & bad elements during good and bad moments.

2. Use the best tool for the job (whether you have to rent, or buy or not). Wireless lavs are important when the subject is mobile and the shot is so WIDE that booming is not optimal, for example. If the subject is at a podium or seated, there may be better alternatives (wired lav, mike on stand, desk-mounted mike). This is also at the mercy of the environment, so elimination of environmental problems or distractions is also paramount.

Please say in more detail how the shot was composed (including distances, other ambient signals, location, equipment model #s, software options, budget, etc), and I could give some more pointers (both for future shoots and to fix the current one).

Scott

[jimmyy]

You could also try replacing the dialogue with new voice talent.

thank you Khaver,
I thought about it, do you think that it's feasible to get the exact speed/pauses with the original video?

[jimmyy]

Thank you Scott,

Could you provide a link that could help me fully understand the basic concepts, volume, frequency, dynamics, phase, formant, spectrum, delay/echo, etc.
Even though I've studied almost all of them in university (it was not my major, but nevertheless I did study engineering) and even though I feel a bit ashamed having forgotten many of the basic concepts.

The problem with true, real-life NOISE is that it is random. This makes it particularly difficult to do something as simple as subtract out (mixing out-of-phase). Plus, even though it is officially called statistically "random", it has local variations that make it uneven to filter against.

Try uploading a 10-30sec clip of each of the 2 tracks in isolation (or if they are separate L+R of a dual track, send them that way and describe them). Please, DO NOT send them pre-mixed. We can try our hand at it, or give suggestions...

I will do that, I've shot in three locations, I'll place some samples of each of the location. I believe that there is only a stereo track recorded by the videocamera, that I can split in left and right.

First, let me give you a few hints about NR:

1. ISOLATION is the name of the game. If you go too extreme, however, you will lose quality in the desired signal along with losing the undesired signal (noise).

I noticed this in audacity when I tried to remove too much the noise at once, I would get very poor result of the useful part of the sound as well.

It is best to do this in MANY passes, if necessary, so that each pass does only minimal damage. Think of "chipping away at block of stone with a scalpel instead of a sledgehammer to get to the bust inside".

2. When you isolate, you must first analyze what all the elements are combining to give you the composite sound that you recorded. Analyze them in ENGINEERING terms: volume, frequency, dynamics, phase, formant, spectrum, delay/echo, etc. See how the signal that you desire compares with the signal you don't. Use those terms to find clearly where there is some wide-enough difference in the demarcation between the good and the bad signal - and then use tools that work on those features.

By analyzing them, I assume that you are referring to a software that does the analysis.

For example: voice has a limited spectrum (~80-6000Hz, less if speaking). If you clearly do not care about other signals beside the voice, a simple voice-optimized static bandpass filter will filter out other frequencies that fall outside of that bandpass. This then makes signals that might be partially in and partially out of that bandwidth be less in volume compared to the (relatively) untouched voice volume. This ratio change can then be taken advantage of in the next stage (like I said, multiple stages).

Sounds great. I need only the voices, nothing else.

Note, that if you DON'T want to (or can't because of logistical reasons) do all passes in realtime in sequence (daisy-chaining), you should ALWAYS use uncompressed/losslessly-compressed files as intermediates.

it sounds very interesting but I'll have do read a bit more about it. I assume that the uncompressed files (based on my limited knowledge) are wav, ogg, flac, whereas the compressed files are mp3...

Some suggestions for the future:
1. You get optimum audio signal quality (in terms of SNR) from these factors (in roughly descending order of importance): Mike Placment/Distance, Mike PolarPattern/Sensitivity, Subject Loudness, Environment, Mike internal & subsystem Quality, and Mike Physical Form-factor.

Your suggestions are more than welcomed.
I realised when I was analysing the results of the shooting that in scene 1 where I had two actors, one with lav mic placed on the collar, and the actor was speaking up, the quality is very good. However in scene 2, another actor, lav placed a bit lower on the blouse cleavage and the actor speaking down (as compared to the first one), the quality is poor.

What also distracted me, is that I was using headphones connected to the camera, while shooting, and you hear much better the sound in the headphones, whereas, at home after having exported the movie, it's not as good...


Note that placement is at the top. You can maybe get GOOD but you likely never get GREAT audio with a shotgun mike that is camera-mounted, because that is NOT the optimal place for the microphone. That's only an optiimal place for the videographer. This is why BOOM poles are so important, becaues you can manually & instantaneously move the mike into the best position possible (within the usual parameters of not being visually noticeable, etc). Lavs are good also, but their form factor often limits their quality & sensitivity. And their placement is very tenuous, having good & bad elements during good and bad moments.

2. Use the best tool for the job (whether you have to rent, or buy or not). Wireless lavs are important when the subject is mobile and the shot is so WIDE that booming is not optimal, for example. If the subject is at a podium or seated, there may be better alternatives (wired lav, mike on stand, desk-mounted mike). This is also at the mercy of the environment, so elimination of environmental problems or distractions is also paramount.

The best tool also costs a lot of money to rent or to buy. I've always searched the best quality/price ratio. On the other hand, thinking now at all the hassle that I had to go trough to shoot the scenes, (find/pay the actors...) I would definetly pay more to have at least a second lav mic, but unfortunatelly they did not have any other on the shooting day. It's a long story, let's say that I was initially planning on using a camcorder, but then when I arrived at the shooting place (in another country, and I arrived one day before the shooting), I realised that my initial estimations were quite far from reaching a good quality in the environment that I was planing to shoot.

Please say in more detail how the shot was composed (including distances, other ambient signals, location, equipment model #s, software options, budget, etc), and I could give some more pointers (both for future shoots and to fix the current one).

I'll get some print-screens of the shots and post them here, it might be much easier this way.

For the equipment I used:
camera: SONY DSR 170
lav mic: Sony UWP C1, connected via XLR to the camera
mic on camera was the one that came with the camera

I've transfered the movies from the mini DV tapes to the computer by unsing firewire and the Windows Movie Maker in wmv format.
The budget I'll have to negotiate with the spouse , I need it only for 5-10 shootings, but to rent all the above including the tripod I paid around 100$ day, so if I have to shoot 7-10 times I'll have to pay 100$. The other issue was that since I had to rent the camera I did not have the time to play with it before...
I'm afraid that even a budget of 1000$ would not cover both the camera and two lav mics which is my target.

It's quite important for me, to have a clear sound, just because I'm shooting lessons where you need to clearly hear the pronounciation. I would say that I'm not searching for perfection, but at least better than average...


I would like to thank you very much for your kind advices, I'm on the learning curve, but this is only a hobby, it will not generate any income for me(the lessons that I want to shoot will be accessible for free), so I'll have to study/learn in the evenings/nights, but I'm definetly willing to invest some time, energy and money.

[jimmyy]

Hi Scott,

I managed to edit and cut a few seconds from each important scene so that you have a sample of each of them.

The first file, Tape1 audio sample.flac was shot in the subway. One actress was in the tickets booth(and had the lav mic) while the second one was outside without any lav mic, so it's the mic mounted on the camera that recorded the voice of the second one. The camera was 2-3 meters far from the actresses. You can hear the noise of the other people entering and exiting the subway.

The second file, Tape 1_1 audio sample.flac was taped in a parc, next to a relatively busy street. One actress had the lav mic on, the other one had no lav mic. I guess that the noise is coming from the wind, and the cars...

The third file Tape 3 audio sample.flac, was shot in a restaurant, and I couldn't get the waiters to turn down completely the music, so you can still hear the background music.

I'm looking forward to receiving your input, especially to learn how to do it better next time.

Please let me know if you need more details, or if I should upload other files. I took the samples by using Audacity and exported the samples in flac format (as I wrote above, first I extracted all the audio(in flac format) of the video by using vlc media player , and then cut only the sample in Audacity and keeping the flac format)

Many thanks for your support

[jimmyy]

What I forgot to mention, is that except for scene 1, for the scene two and three, if I split the flac from stereo into left and right and play only the lav mic, it sounds like it's better, since the two actresses were close one to the other (aprox 1m) and the lav mic captured both of them.
Shall I do that? rely only on the lav mic? for the second and third files?

[Cornucopia]

I'll listen to those clips tonight and get back to you tomorrow...

Scott

edit: 1/25/13, still working on it..

[jimmyy]

Hi,

While waiting for Scott to have a look at the samples that I posted, I have one additional question.

The recordings that I've made are on two separate channels, and while listening to it on the headphones you can hear in the left hedphone one voice and in the right one the other person replying.

How could I merge them so that you can hear in both headphones both channels. I guess that I'll have do to some sort of overlap....


Thank you

[aedipuss]

here's what i would do. compress and normalize the channels individually. combine into mono. convert back to stereo.

you need the background noise or else the audio will be too weird for people who are filmed in public.