Dear Temple Of Goo readers,
Today’s post is aimed at clarifying a concept which has been inquired about quite a few times both via this blog and through our Official YouTube channel.
During the past few weeks I’ve gotten the following question, or variations thereof:
“Where did you find those instrumental Goo Goo Dolls songs?“
If you’re not sure what some of our readers / YouTube watchers are referring to, I would encourage you to visit our aforementioned (and aforelinked, heh) YouTube channel and delve into it, locating videos like these two:
The instrumentals we post on your YouTube channel are not official ones, that is officially released by the band. They are AI-generated using the latest available technologies at our disposal, unless specified otherwise (keep reading).
Without diving too much into the technology itself and its multifaceted nature, which is at its current stage still quite complicated and not easily utilizable by everyone, I will try to explain it in the most efficient and easy-to-understand way. If you prefer the opinion of an expert on the matter, I encourage you to check out this greatly-written blog post. Another amazing resource is the Audio Separation Discord Server, however I would suggest this only to those who are willing to get into the more technical aspects of AI separation.
These AI programs, or rather models, are trained using a plethora of songs from different eras and genres. They are fed entire catalogs of music in order to learn how to recognize each construct comprising a song, such as vocals, drums, guitars, etc… and are developed so as to extract those components (which, in technical terminology, are referred to as “stems”), then reconstructing the song without a chosen part. In the following video, where I launched the “Goo Goo Dolls Stem Mix” project, a mathematical explanation is provided as well. I encourage you to watch it if you seek a very basic idea of what the AI technology is capable of, and how it does operate.
While a lot of combinations are possible, as explained by the clip above, the most common one is actually the vocals-less version, that is the instrumental counterpart of a given tune.
Consequently, that is why you will find a lot of instrumental Goo Goo Dolls songs on our channel, which we create by using the latest AI models to extract the vocals stem. In some cases it is also possible to do such operation with live takes, provided the recording comes with high quality.
If you want to play with this amazing AI technology yourself, there are different ways to do so. Obliviously, Google is your friend if you feel like doing your own research. You can Google for terms like “stem separation” or “audio separation” to get you started. You will find a lot of pages talking about it, and along with them, different softwares (most of them Python-based, no GUI) and some websites to play around with the AI models.
The easiest way to test and play around with the technology yourself is by visiting Mvsep, a fantastic and user-friendly resource, uploading a song and choosing a model for AI processing. There are several models you can choose from. As of now, I would recommend choosing the MDX23C one, utilizing the variation with the highest SDR value, that is this one:
Important: For optimal results, you should use a lossless source (such as FLAC or WAV files), rather than an MP3 one. AI models operate way better with lossless files, rather than lossy ones, which are too compressed for good stem separation.
Since the technology is still being (heavily) developed, sometimes results may be decent, if not very scarce. Other times, they are almost perfect. Most of the times they tend to be good / very good, depending on the chosen model(s) and on the original state of the song itself (its “production value”).
Like I stated above, there are plenty of different models (and even combinations of models where two or more of them are used in sequence during AI processing) and there is still no consensus on which model/combo is the best one. There is, however, a way to “evaluate” the degree to which they somewhat get closer to the real version of the instrumental counterpart, that is by measuring something called SDR. The higher the SDR value of a processed song, the closer the AI-generated instrumental song is similar to its real counterpart.
YOU SHOULD BE HAPPY EP (2017) – THE ORIGINAL INSTRUMENTALS
Speaking of real instrumentals, to my knowledge the Goo Goo Dolls never released official instrumental versions of their songs, with the only exception being the 2002’s single “I’m Still Here“, which was a John Rzeznik-only song, by the way.
In reality, there are original (real, non-AI-generated) Goo Goo Dolls songs currently commercialized but only to a very small audience, that is DJ’s. Yes, you read it correctly. Some USA DJ’s are given access to a special platform where they can buy original instrumental songs, including some Goo Goo Dolls ones, to use during their music sessions / concerts.
Which instrumental albums do those DJ’s have access to? “Magnetic”, “Miracle Pill”, “Boxes”, and the one we’re sharing with you today: “You Should Be Happy EP”. I’m not sure about the rest of the discography, those are the ones I know they can get for sure.
Below you can listen to the “You Should Be Happy EP” Original Instrumental songs, hence non-AI-generated. These songs were given to me by one of those DJ’s a while ago, and it’s now time to share them with you. Given the fact they are commercialized, I cannot provide a download link for them, and can only make them available as streamable content on YouTube. I hope you will enjoy them, and for those who are interested in the AI technology I briefly discussed above, I hope you will do your own research about it and eventually discover many amazing things like I’ve been discovering them during the past couple of years since those AI models began being publicly available.
TRACK LISTING:
- Tattered Edge / You Should Be Happy
- Use Me
- 30K Feet
- Walk Away
- Boxes (Remix)
Temple of Goo - The rarest Goo Goo Dolls releases, for real Goo fans only 
Ok that there you are one sounds like the remixed version that appears in ego opinion art and commerce! I’ve been looking for more digital versions of those. Do you happen to have a mp3 of the re-recorded two days in February?