Better speech compression with Opus

In a previous post, we looked at configuring Audex for ripping CDs to the superior Opus format. While that was good for general purpose audio encoding, it doesn’t let us take advantage of Opus’ immense versatility, it’s ability to get excellent results across a wide range of bit rates.

You will notice at the low end of that graph are codecs that typically specialize in speech, because speech is more compressible that complex audio sources like music. Speex, for instance, can get excellent compression of speech with acceptable quality, yielding much smaller file size than MP3 or Vorbis. One of the benefits of Opus is that it should be able to replace this usage as well.

As you might have guessed, the opusenc commands we gave Audex can also be used directly on the command line. If you want to target a lower bit rate for speech encoding, all you need to do is add the --bitrate option. For speech, try something in the range of 16-48. In Audex, you might add two different profiles for compressing speech with Opus, just as there are multiple profiles for different qualities of compression with Vorbis.

opusenc takes raw, Wave, AIFF, or FLAC input (FLAC is still the preferred format for archival storage). If you pass a FLAC file that is already tagged with metadata, opusenc will incorporate that metadata automatically, so there is no need to pass it on the command line.

opusenc --bitrate 16 input.flac output.opus

For comparison, a 40 minute audio recording ripped as an uncompressed Wave file was about 403 MB, compressed with FLAC it was about 103 MB, opusenc with default settings got it down to about 26 MB, while the above command reduced it to about 4.5 MB. So targeting the lower bit rate was a space savings of over 80% and made it small enough to share by email.

Ripping CD to .opus with Audex on Debian

The Opus codec is basically the one codec to rule them all. It is an open standard (RFC 6716) and it outperforms almost all other codecs, open and proprietary, over a huge range of bitrates. It also has a very low latency which is important for internet applications involving real-time communications and it is already supported by modern browsers like Firefox, Chrome, and Opera. But if you want to start storing your music in the best available format, you might have difficulty finding convenient tools that support it. Well, here is one that worked for me, so I’ll share it. It was pretty simple, although not entirely obvious.

Audex is a graphical CD ripper for KDE. It does not support ripping to Opus out-of-box, but it does have a feature to add a new encoder. For this, we will just use the command-line encoder found in opus-tools, so install the packages audex and opus-tools. Now, when you run Audex, go to “Settings” > “Configure Audex...”, select the “Profiles” section, and then “Add...”. Name the new profile “Opus”. Under the “Encoder” tab, select “Custom” from the first drop down. Under “Command pattern:”, put in “opusenc $i $o”, and under “Suffix:” put in “opus”. You can look at some of the other options if you wish (I set mine to replace spaces with underscores), but that is all that is necessary to get it working. Save your new profile, select the profile from the main window, and begin ripping your favorite discs.

If you also want to capture metadata, you will need a newer version of opus-tools than what is in Wheezy. You can download the source, and build it yourself. You will need to have the wheezy-backports repository enabled to get some of the dependencies:

sudo aptitude -t wheezy-backports install libogg-dev libflac-dev libopus-dev

Once the dependencies are met, you should be able to configure and install opus-tools:

cd opus-tools-src-dir
./configure
make
sudo make install

Now, you can change the “Command pattern:” in Audex to “opusenc --title "$ttitle" --artist "$tartist" --album "$title" --date "date" --genre "$genre" $i $o”. This will capture metadata and embed it in your new Opus files.

Update: If you are looking to get better compression for speech recordings, such as lectures, sermons, speeches, etc, see this post.