I want the loudest peak sound in a movie clip to be as loud as the codec allows, then have every other sound amplified accordingly.
What's a practical example in order to accomplish this using ffmpeg?
Super User is a question and answer site for computer enthusiasts and power users. It only takes a minute to sign up.
Sign up to join this communityI want the loudest peak sound in a movie clip to be as loud as the codec allows, then have every other sound amplified accordingly.
What's a practical example in order to accomplish this using ffmpeg?
Current ffmpeg has three filters that can be directly used for normalization – although they are already quite advanced, so they do not simply apply gain to reach a peak level. If you want a “simple” RMS-based or peak normalization to 0 dBFS (or any other target), go to the last section of this answer.
The basic usage for a filter is:
ffmpeg -i input -filter:a filter output
Here are the three more advanced filters:
loudnorm
: loudness normalization according to EBU R128. You can set an integrated loudness target, a loudness range target, or maximum true peak. This is recommended for publishing audio and video and it is used by broadcasters all over the world. You should use this filter in a two-pass mode, so I recommend my wrapper script ffmpeg-normalize
for it (for more info see below).
dynaudnorm
: “intelligent” loudness normalization without clipping, which applies normalization dynamically over windowed portions of the file. This may change the characteristics of the sound, so it should be applied with caution. Example for medium-level compression:
ffmpeg -i input.wav -filter:a "dynaudnorm=p=0.9:s=5" output.wav
speechnorm
: specifically made for speech normalization. See the examples here.
Also, the volume
filter can be used to perform simple volume adjustments. See the Audio Volume Manipulation wiki entry for more.
ffmpeg-normalize
toolSince you should use the loudnorm filter with two passes, and that might be a bit complicated to wrap in a script, I created a Python program to normalize media files, available on PyPi as well. You simply:
ffmpeg
executable in your $PATH
by either adding it in, for example, /usr/local/bin
, or adding its directory to $PATH
pip install ffmpeg-normalize
ffmpeg-normalize
For example:
ffmpeg-normalize input.mp4 -o output.mp4 -c:a aac -b:a 192k
Or, to simply batch-normalize a number of audio files and write them as uncompressed WAV to an output folder:
ffmpeg-normalize *.m4a -of /path/to/outputFolder -ext wav
The tool supports EBU R128 (default), RMS and peak. Have a look at ffmpeg-normalize -h
for more options and check the README for some examples.
Also, it supports re-encoding with other encoders (e.g., AAC or MP3), or automatic merging of the audio back into the video.
ffmpeg
In ffmpeg you can use the volume
filter to change the volume of a track. Make sure you download a recent version of the program.
This guide is for peak normalization, meaning that it will make the loudest part in the file sit at 0 dB instead of something lower. There is also RMS-based normalization which tries to make the average loudness the same across multiple files. To do that, do not try to push the maximum volume to 0 dB, but the mean volume to the dB level of choice (e.g. -26 dB).
First you need to analyze the audio stream for the maximum volume to see if normalizing would even pay off:
ffmpeg -i video.avi -af "volumedetect" -vn -sn -dn -f null /dev/null
Replace /dev/null
with NUL
on Windows.
The -vn
, -sn
, and -dn
arguments instruct ffmpeg to ignore non-audio streams during this analysis. This drastically speeds up the analysis.
This will output something like the following:
[Parsed_volumedetect_0 @ 0x7f8ba1c121a0] mean_volume: -16.0 dB
[Parsed_volumedetect_0 @ 0x7f8ba1c121a0] max_volume: -5.0 dB
[Parsed_volumedetect_0 @ 0x7f8ba1c121a0] histogram_0db: 87861
As you can see, our maximum volume is -5.0 dB, so we can apply 5 dB gain. If you get a value of 0 dB, then you don't need to normalize the audio.
Now we apply the volume
filter to an audio file. Note that applying the filter means we will have to re-encode the audio stream. What codec you want for audio depends on the original format, of course. Here are some examples:
Plain audio file: Just encode the file with whatever encoder you need:
ffmpeg -i input.wav -af "volume=5dB" output.mp3
Your options are very broad, of course.
AVI format: Usually there's MP3 audio with video that comes in an AVI container:
ffmpeg -i video.avi -af "volume=5dB" -c:v copy -c:a libmp3lame -q:a 2 output.avi
Here we chose quality level 2. Values range from 0–9 and lower means better. Check the MP3 VBR guide for more info on setting the quality. You can also set a fixed bitrate with -b:a 192k
, for example.
MP4 format: With an MP4 container, you will typically find AAC audio. We can use ffmpeg's build-in AAC encoder.
ffmpeg -i video.mp4 -af "volume=5dB" -c:v copy -c:a aac -b:a 192k output.mp4
Here you can also use other AAC encoders. Some of them support VBR, too. See this answer and the AAC encoding guide for some tips.
In the above examples, the video stream will be copied over using -c:v copy
. If there are subtitles in your input file, or multiple video streams, use the option -map 0
before the output filename.
loudnorm
(or other) filter: ffmpeg -i input.wav -filter:a loudnorm output.wav
Here's a script to normalize sound levels of .m4a files. Watch out if the sound levels are too quiet to start with. The final sound can be better if you use something like Audacity in that case.
#!/bin/bash
# Purpose: Use ffmpeg to normalize .m4a audio files to bring them up to max volume, if they at first have negative db volume. Doesn't process them if not. Keeps bitrate same as source files.
# Parameters: $1 should be the name of the directory containing input .m4a files.
# $2 should be the output directory.
INPUTDIR=$1
OUTPUTDIR=$2
<<"COMMENT"
# For ffmpeg arguments http://superuser.com/questions/323119/how-can-i-normalize-audio-using-ffmpeg
# and
# https://kdecherf.com/blog/2012/01/14/ffmpeg-converting-m4a-files-to-mp3-with-the-same-bitrate/
ffmpeg -i test.m4a -af "volumedetect" -f null /dev/null
ffmpeg -i test.m4a -af "volumedetect" -f null /dev/null 2>&1 | grep max_volume
# output: max_volume: -10.3 dB
ffmpeg -i test.m4a -af "volumedetect" -f null /dev/null 2>&1 | grep 'max_volume\|Duration'
# Output:
# Duration: 00:00:02.14, start: 0.000000, bitrate: 176 kb/s
# [Parsed_volumedetect_0 @ 0x7f8531e011a0] max_volume: -10.3 dB
ffmpeg -i test.m4a -af "volumedetect" -f null /dev/null 2>&1 | grep max_volume | awk -F': ' '{print $2}' | cut -d' ' -f1
# Output: -10.3
ffmpeg -i test.m4a 2>&1 | grep Audio
# output: Stream #0:0(und): Audio: aac (LC) (mp4a / 0x6134706D), 44100 Hz, stereo, fltp, 170 kb/s (default)
ffmpeg -i test.m4a 2>&1 | grep Audio | awk -F', ' '{print $5}' | cut -d' ' -f1
# output: 170
# This works, but I get a much smaller output file. The sound levels do appear normalized.
ffmpeg -i test.m4a -af "volume=10.3dB" -c:v copy -c:a aac -strict experimental output.m4a
# Operates quietly.
ffmpeg -i test.m4a -af "volume=10.3dB" -c:v copy -c:a aac -strict experimental -b:a 192k output.m4a -loglevel quiet
COMMENT
# $1 (first param) should be the name of a .m4a input file, with .m4a extension
# $2 should be name of output file, with extension
function normalizeAudioFile {
INPUTFILE=$1
OUTPUTFILE=$2
DBLEVEL=`ffmpeg -i ${INPUTFILE} -af "volumedetect" -f null /dev/null 2>&1 | grep max_volume | awk -F': ' '{print $2}' | cut -d' ' -f1`
# We're only going to increase db level if max volume has negative db level.
# Bash doesn't do floating comparison directly
COMPRESULT=`echo ${DBLEVEL}'<'0 | bc -l`
if [ ${COMPRESULT} -eq 1 ]; then
DBLEVEL=`echo "-(${DBLEVEL})" | bc -l`
BITRATE=`ffmpeg -i ${INPUTFILE} 2>&1 | grep Audio | awk -F', ' '{print $5}' | cut -d' ' -f1`
# echo $DBLEVEL
# echo $BITRATE
ffmpeg -i ${INPUTFILE} -af "volume=${DBLEVEL}dB" -c:v copy -c:a aac -strict experimental -b:a ${BITRATE}k ${OUTPUTFILE} -loglevel quiet
else
echo "Already at max db level:" $DBLEVEL "just copying exact file"
cp ${INPUTFILE} ${OUTPUTFILE}
fi
}
for inputFilePath in ${INPUTDIR}/*; do
inputFile=$(basename $inputFilePath)
echo "Processing input file: " $inputFile
outputFilePath=${OUTPUTDIR}/$inputFile
normalizeAudioFile ${inputFilePath} ${outputFilePath}
done
I can not comment on the best message so that is my ugly bash based on it to do that
ffmpeg -i sound.mp3 -af volumedetect -f null -y nul &> original.txt
grep "max_volume" original.txt > original1.tmp
sed -i 's|: -|=|' original1.tmp
if [ $? = 0 ]
then
sed -i 's| |\r\n|' original.tmp
sed -i 's| |\r\n|' original.tmp
sed -i 's| |\r\n|' original.tmp
sed -i 's| |\r\n|' original.tmp
grep "max_volume" original1.tmp > original2.tmp
sed -i 's|max_volume=||' original2.tmp
yourscriptvar=$(cat "./original2.tmp")dB
rm result.mp3
ffmpeg -i sound.mp3 -af "volume=$yourscriptvar" result.mp3
ffmpeg -i result.mp3 -af volumedetect -f null -y nul &> result.txt
fi
I'd like to offer my own ffmpeg-based solution NormaWave for people who are new to ffmpeg and seeking alternatives to heavy and inaccurate software (Adobe Audition in my case). After setting up only 7 parameters, you have to click a single button, select the files and wait. It is as simple as possible. The script will read stats and apply up to 9 passes to some of the files to obtain your desired parameters (integrated loudness, loudness ratio, and maximum true peak).
You'll need Excel 2016 or later. It has the user-friendly interface. There are built-in instructions, links to the latest editions of ffmpeg and FLAC, and entry fields' descriptions with bits of theoretical information.
Select your input and output files' extensions (WAV, FLAC, MP3, or WEBM).
Type in your parameters (IL, LRA, and TP) and the sample rate. The built-in check in every field won't let you input ffmpeg-unacceptable values. Or click "Delete/Backspace" for default values.
In the beginning and in the end your files will be renamed back and forth to prevent ffmpeg failure.
When job is done, you can analyze *.bat files and *.txt reports (created in the working folder) to decide what you should do with the final results.
No matter what filter you choose, the first pass is always the reading stats with loudnorm. There may be up to 9 passes, depending on your desired output parameters. Decreasing LRA is a very difficult task.
I use command line (adjust it by yours demands)
mkdir NORMALIZED; for mkv in *.mkv ; do wav=`basename "$mkv" .mkv`.wav; echo "${wav}"; ffmpeg -i "${mkv}" -acodec copy "${wav}" ; normalize-audio "${wav}"; ffmpeg -i "${mkv}" -i "${wav}" -codec copy NORMALIZED/"${mkv}"; done;
Or batch
[ ! -d work ] && echo "work directory does not ezists" && exit 1
for fn in *.mkv
do
map=$(ffmpeg -i "$fn" |& grep Stream | grep pcm_s16le) # pcm_dvd ac3 mp3
if [ "$map" != "" ]
then
echo "$fn"
fb=${fn%.mkv} #; shopt -s extglob; fb=${fb%%+([[:space:]])}
amap=${map:12:3}
ffmpeg -y -i "$fn" -map $amap -acodec pcm_s16le work/"$fb".wav 2>/dev/null
wavegain -y work/"$fb".wav
ffmpeg -y -i "$fn" -i work/"$fb".wav -c:v copy -c:a copy -map 0:v:0 -map 1:a:0 work/"$fb".mkv 2>/dev/null
rm -f work/"$fb".wav
echo
fi
done
read -p "BAIGTA"
ffmpeg -i image.jpg -i "input.mp3" -acodec copy tmp.avi
mencoder -ovc copy -oac copy tmp.avi -of rawaudio -af volnorm=1 -oac mp3lame -lameopts cbr:preset=192 -srate 48000 -o "output.mp3"
rm -f tmp.avi