I’m using Audacity 1.3.12 in CleanSpeech mode to edit sermons for my church’s website. However, even after applying CleanSpeech, I still spend a lot of time editing the recordings because many parts of the sermon are not loud enough. The people making the original recordings do not always use the correct volume, on top of the fact that the pastor is loud at times and soft at other times. While we can hear everthing fine in person, I’m not sure the audio files are that clear unless I use “leveller” to make the soft parts louder. I am not an audio engineer and don’t know nearly enough to come up with my own presets or batch processing steps using the regular (non-CleanSpeech) mode.
I did find how to edit the CleanSpeech presets under “Edit Chains,” but I did not see anything that I could change to help get the soft parts louder without affecting the parts that are already loud. I see the “Normalize” step listed twice with “ApplyGain=yes RemoveDCOffset=yes Level=0.000000” each time, and a step for “Leveller” saying “dB=-70dB Passes=2.” I don’t see anything on how much “gain” can be applied. Will changing the dB for Leveller help? Or will it amplify the loud parts and cause clipping or other distortion? Or is there a way to change the default “gain” that is applied when Normalizing? Or can someone suggest other batch steps which I could import into the regular Audacity mode?
I think you have misunderstood what CleanSpeach is really for: it is not an effect to be applied to clean up speach recordings and make them clearere - rather it is a cut-down version of Audacity with a simpler interface, not giving access to many useful tools.
If I were you I would turn off CleanSpeach and use the “proper” Audacity (in the latest Alpha version 1.3.14 CleanSpeach has been removed by the developers as it causes far too many problems).
You need to use a Compressor for this - there is a Compressor effect buitl in to Audacity: select the audio and use Effect > Compressor.
There is a better compressor available called Chris’ Dynamic Compressor which can be used as an Audacity plug-in. See this thread: https://forum.audacityteam.org/t/chris-capels-dynamic-compressor/20241/1
Oh and upgrade to the latest Beta 1.3.13 which you can get from here: http://audacityteam.org/download/
The “Clean Speech” chain is a quick and dirty method to process speech recordings to make the volume more level and hopefully more intelligible and output the processed audio as an MP3 file (with an additional MP3 output after the initial “Normalize” step but before any other processing. It is not a high quality effect but is a quick and easy “one click solution” that can be useful for processing speech recordings quickly with minimal effort.
Here’s what the chain does:
- Converts stereo tracks to mono
- The “Normalize” steps in the chain amplify the audio as much as possible without clipping the loudest parts of the recording.
- SaveHqMaster1: This exports an MP3 copy - the MP3 compression settings are whatever settings were used when MP3 export was last used.
- Noise Removal effect - Uses whatever settings were used the last time this effect was last used.
- TruncateSilence: shortens long silences within the recording
- Leveller: evens out the volume differences between loud and quiet sections
- Normalize (same as step 2)
- Makes final MP3 Export - the MP3 compression settings are whatever settings were used when MP3 export was last used.
To even out the volume more than currently happens, you could increase the “strength” of the Leveller effect (Degree of levelling), but note that this effect is rather brutal. The Leveller effect “squashes” the waveform where the waveform exceeds the “Threshold” level. This will introduce a degree of distortion, but not as bad as if the loud parts were clipped. The default setting for Leveller in the Clean Speech chain is “Moderate” (2 passes) which will even out the volume level by a “moderate” amount and will cause a “moderate” amount of distortion to the loud parts. You could increase this setting to “Heavy, Heavier, or Heaviest” to increase the strength of this effect.
Alternatively you could add the “Compressor” effect into the chain (probably immediately before the Leveller effect).
To work out the best settings you would need to manually run each of the chain steps manually on a “typical” recording, then when you’ve worked out the optimum settings, make a new Chain with those settings.
As a starting point I’d suggest adding the Compressor effect immediately before the Leveller effect with these settings:
Noise Floor: -50
Make-up gain: Selected (On)
Compress based on Peaks: Selected (On)
These are aggressive settings that should make a significant difference.
(Oh yes… and upgrade to 1.3.13)
Thank you so much for your help. I tried using the compressor with settings similar to what you suggested. I used a Noise Floor of 40, as it sounded better on the preview, and I decided to try the Ratio at 9:1 instead of all the way to 10:1. This worked a lot better and I did not need to do any manual editing to the volume of this recording. I may need to experiment with the noise floor some more, as this did add noise to the recording and I had to use the noise removal effect. It would be nice to have an explanation, in simple terms that beginners can understand, of what the numbers for each of these setting options mean and what they do.
If the developers change their mind and decide to bring back a new “clean speech” mode in future versions of Audacity, I think they should add an aggressive “compressor” setting to the chain. This seemed to work much better to amplify and even out the volume than “normalize” and “leveler.” Or perhaps it depends on the quality of the original recording; if so, they could provide an option to chose the optimal chain that matches the general quality of the original file.
Thank you again for your help.
It’s not actually adding noise, it’s making noise that is already there louder and so more noticeable. To avoid this, try setting the noise floor higher (a smaller number, for example -30 rather than -40 dB). The compressor will amplify low level sound that is above the noise floor, but does not amplify noise that is below the noise floor.
The important thing to remember when working with dB units is that “0 dB” is maximum (full scale). As sounds get quieter the dB level is increasingly negative, so:
0 dB = maximum
-6 dB = half track height (fairly loud)
-20 dB = fairly small and fairly quiet
-80 dB = almost flat line and very very quiet.