Mastering with LANDR v.s. Mastering Myself

LANDR has been around for a while now, and I finally decided to try it out for myself. I recently mastered a track I produced a while ago, a bad idea, I know. I put the track to rest for a month, so I could master it with fresh ears because I didn’t have the budget to hire a mastering engineer. It worked out all right, but I was curious how a service like LANDR would handle it.

So I signed up for a free account on LANDR.com and uploaded my track. It’s an electronic progressive house track which I mixed with plenty of headroom (10dB) left for mastering. LANDR allows you to master 2 tracks for free each month. The free version generates a 192kbps mp3 file and doesn’t provide you with extra mastering options, but it’s a good way to test out the service.

I was surprised

I have to say I was surprised by the result! It did a much better job than I initially thought was possible for an automated mastering service. The LANDR version had much more impact and low end than my own master. It really put an extra layer of quality on top and I was actually disappointed about the quality of my own self-mastered version.

The LANDR master had more punch and clarity in the high end and when comparing the two versions I noticed my own master had some muddy frequencies left that were gone in LANDR’s version. So in terms of quality I think it did a great job. And I only tested this with the free version.

LANDR has 3 different paid plans that give you access to additional features. In addition to this monthly or annual fee you still have to pay per track, but this amount is discounted according to the plan you have. The Pro plan, however, allows you to use unlimited high quality, 24-bit Wav files, without paying per individual track. I will review the paid plans in a feature article.

So how does it work?

I have heard about LANDR for a while now and since the increasing interest in artificial intelligence and machine learning an automated mastering service was inevitable. The concept of machine learning is pretty simple. You create an algorithm that analyses lots of tracks in one specific genre. Data such as, loudness, tonal balance and stereo width can be easily measured. Thousands of tracks are analysed and an average is created that is applied to those tracks you upload.

The hard part is programming an algorithm that works with every possible mix without human interaction. I actually asked LANDR if they could provide me with some information about the algorithm and they were kind enough to respond with the following:

“LANDR has been developed and improved over the years to understand how to best treat your music by making subtle mastering decisions that are appropriate for the genre of your track and by identifying audio features that are most important. Our AI identifies the genre-specific features and compares them across a large number of tracks, making decisions about EQ, stereo width, compression, tone, saturation, loudness, that will make it stand out and be perfect for the style of music. These changes can be small or big, depending on your mix. We also have identified variations in aesthetics for each genre that can alter the feel and emotional impact, with our mastering styles preferences.”

LANDR.com

Technical Comparison

Loudness

My unmastered track peaked at -10.07dB while my integrated loudness was around -26.3 LUFS. To learn more about LUFS check out this video by mastering engineer Ian Shepherd.

Because Spotify normalizes music to -14 LUFS, anything louder than that wouldn’t make much sense as it will be turned down automatically. This loudness standard is used to bring an end to the loudness war. My self-mastered track ended up at -13.3 LUFS which helped preserve some dynamics. The master by LANDR ended up at -10.7 LUFS which is a little loud for my taste but it’s not too bad, considering we’re talking about electronic music here.

If we take a look at the true peak level, however, there is a problem. We all know that 0dBFS is the limit of digital audio. Nasty clipping occurs if you cross this threshold. But even though your peak meter might be displaying 0dBFS, clipping can still occur. This is because digital audio needs to be converted back to analog to be able to hear it through speakers or headphones. During this DA conversion process, filters are used to turn a stepped digital wave into a smooth analog one. This process takes up headroom, however, which can cause inter-sample peaks. Check out this article by protools expert to learn more about true peak levels.

High-quality DA converters might be able to handle this problem, but most consumers listen to music on their phones and laptops which often have cheap converters. Spotify has the following recommendations about loudness and true-peak levels:

Mastering tips for Spotify

“Target the loudness level of your master at -14 dB integrated LUFS and keep it below -1 dB TP (True Peak) max. This is best for the lossy formats we use (Ogg/Vorbis and AAC) and will ensure no extra distortion is introduced in the transcoding process.

If your master is louder than -14 dB integrated LUFS, make sure it stays below -2 dB TP (True Peak) max to avoid extra distortion. This is because louder tracks are more susceptible to extra distortion in the transcoding process.”

As found on the spotify site itself.

My master is a little louder than -14 LUFS and should technically have a lower true peak level to account for the conversion process. But because most LUFS meters have an accuracy of +/- 1dB this is acceptable. The LANDR master, however, peaks at +0.84dB while having a loudness of -10.7 LUFS thus having a true peak level almost 3dB higher than recommended. A possible culprit may lie in the mp3 conversion process.

I contacted LANDR about this and they recommended using a LOW-intensity level. There are 3 loudness intensities you can choose: low, medium, and high. I thought this was a premium-only option at first but you can change these settings with the free version as well. The low-intensity version turned out to have an integrated loudness level of -12.9LUFS (a tiny bit louder than my own master) with a true peak level of +0.56dB.

This true peak level is still above the -2dB that Spotify recommends. LANDR’s recommendation was to simply use the master that sounds best for the music. They did tell me they are currently working on additional options to be more compliant with streaming and encoding standards. That’s a good decision because I don’t think streaming will go away anytime soon.

Just because I was curious about true peak levels of other professionally mastered tracks I compared one track that was mastered a while back for a CD release and one track that was recently released for digital media. You can see the difference even by looking at the waveforms. The CD version peaked at 0.00dB with true peaks levels at +1,02dB. The digital release peaked at -0.40dB with true peak levels at -0,26dB.

I do like to be on the safe side myself, but to be honest, I have never heard the artifacts produced by positive true peak levels. Most modern DA converters can probably handle these true peak levels perfectly fine. And just like everything in audio. If it sounds good it is good.

Stereo Width

In my self-mastered version, I used a stereo enhancer to widen the stereo field a bit. By looking at the correlation meter in the free SPAN analyzer my version provided a little bit more width than LANDR’s version. I don’t think they actually enhanced the width because it was similar to the original version. This probably depends on the genre, electronic music doesn’t have much stereo width anyways so the algorithm didn’t find it necessary to increase the width.

Tone

LANDR clearly boosted the low end to give more power to the kick and bass with an additional boost in the higher frequency area. By comparing the spectrum plots in Fabfilter’s Pro Q3 at equal loudness levels I didn’t see big changes, however. I used the peak freeze mode in the image below, which indicates static EQ differences. The LANDR version is indicated by the red line. You can see a slight dip in the mid-band and a rather steep, almost brick wall-like, low pass filter at 19kHz. This steep filter at 19kHz is most likely due to the mp3 format. That’s one of the reasons not to use mp3’s folks!

When we turn off peak-hold and change the spectrum plot response to fast, we can see that there is definitely some dynamic EQ action being used here. The kick is being boosted when played and attenuated in between, the same is true for the high hats. This adds punch without unnecessary clutter.

Compression

My mix was pretty dynamic, so some compression is needed to achieve higher loudness levels. You can see the difference in the images below. Compression is used to enhance transients, as you can see in the low-intensity version. The peaks are louder than in my self-mastered track which results in a more defined attack to the drum sound. The medium intensity master does use quite a bit of limiting, however.

Would I use LANDR for all my tracks?

I really am impressed by the quality level that LANDR produces. And I am definitely interested in using this platform more often especially now that I’m releasing more songs for various projects. I do have to test the platform with other genres though.

LANDR asks €17.99 for a high-quality 24-bit WAV version per track in addition to the advanced subscription. If you were to release 4 tracks a year this would cost you €180,-. This is actually the same price you would pay for a decent mastering engineer which asks €45 per track. If I were to release one EP per year I would rather go to a mastering engineer.

But since I produce music for various media such as film, advertising, ghost production it may be better to go for the Pro subscription which costs €299 per year and allows me to master unlimited high-quality Wavs with no additional cost per track.  I would only need to release 7 tracks per year to save money on a real mastering engineer so for me this subscription makes sense. But this all depends on your individual needs of course.

Do I ever need a real mastering engineer again?

While AI mastering has its benefits like price, and ease of use, I wouldn’t say mastering engineers need to worry about losing their job. There are still lots of reasons to hire a professional mastering engineer.

LANDR currently only works on singles which is fine for today’s streaming market, but if you need to master a whole album with relative loudness levels between tracks including fades, pauses, and markers then AI mastering still has a long way to go. A real mastering engineer would do a quality pass, checking for noises, pops, and clicks that need to be fixed with specialized tools. Professional mastering engineers can provide you with the right files for cd, DVD, streaming, vinyl, surround, and any other formats that you wish to distribute your music in.

Mastering engineers use analog equipment and make decisions on emotions instead of algorithms. The danger in using algorithms is that it can result in average sounding masters. A mastering engineer can think out of the box which a computer program can not. You can provide a mastering engineer with feedback and he can give you tips to improve your mix in return. You can attend a mastering session and listen to your track in an acoustically treated room with high-quality monitors. You’ll learn a lot by just visiting a mastering engineer from time to time.

In Summary

Mastering yourself is something you want to prevent at all times. Even mixing your own tracks can be hard, because you lose objectivity the longer you work on a project. LANDR is a good alternative for mastering your own tracks with surprisingly good results.

I would still go to a real mastering engineer to master my EP’s or Albums, but for single releases, LANDR is a great solution!