Welcome to the brave new world of DSP! Along with folks like iZotope, Audionamix is on the cutting edge of spectral based DSP technology that can do previously unheard of things with audio. These products are both based on being able to take a stereo mix of a song and separate out the vocals from the music, and even separate the music from the drums. There are two basic uses of this kind of product, one creative and one forensic, so in general this will appeal to two different types of users. Copyright infringement aside, being able to pull stems from music will be super useful to people working in electronic, pop, and hip-hop genres. Being able to pull out a vocal from a snippet of music and then sample and loop it could be huge for those producers. The second, more forensic category, is for those folks who hope they'll never need this software but then something comes up and they do need it. I'm in that category. I recently helped produce a live recording of a band at The Fillmore in San Francisco. To make a long story short, the band's crew shorted out our split in between sound check and the show, so we lost the vocal feed on the first two songs of the multitrack. We did however have the two-track mix from the front of house console. After meeting with the nice folks at Audionamix during AES, this seemed like a good test for their software and services.
I started with Trax Pro SP because I assumed Pro means better – right? Well yes and no it turns out. Trax Pro SP is Audionamix's legacy product that's most advanced in terms of options. But more options mean more complexity and more skill needed by the operator to extract the vocals. As I was only trying to help out the producer on this project and not mixing it, Trax Pro SP left too many decisions in my hands that I wasn't comfortable making, nor did I have the time to learn the software in enough depth to really get the best results. The results I got using the automatic extraction were frankly not that good to the point of really not being useful. I wondered if I was just doing something wrong or missing something, so I called Ellie McNeil and Steve Oliver at Audionamix's offices, and they were nice enough to explain to me that, as I've summarized above, Trax Pro SP was going to take me some time to learn and I'd have to make decisions best left to someone else with more skill level or who was directly mixing the project. But Ellie told me that their newer, entry level product XTRAX STEMS used the same technology but also used a machine learning component that might yield better results with an automatic extraction. Their proprietary algorithms are based around artificial intelligence in the form of deep neural networks. Ellie went on to explain; "Audionamix has a team of researchers dedicated to developing new artificial neural networks fine-tuned specifically for vocal source separation. From an ever expanding and diverse aural library of noise, background music, and vocal content, the algorithms are continuously trained to become more adept at separating vocals, be it speech or singing voice, from any other complex soundscape." So, my next step was to try XTRAX STEMS on the same tracks, and I'm happy to report that the results from this program sounded much better and were pretty much usable. The producer who was overseeing the project was largely impressed with the quality of the vocal extraction, so if my client is happy, I'm happy! Plus, XTRAX STEMS is significantly more affordable than Trax Pro SP. I should note that XTRAX STEMS requires an internet connection to operate, and can take some time to process the tracks as it's essentially communicating in a bi-directional protocol with Audionamix's servers to upload, analyze, and download the audio and stems. This is promising, because as the Audionamix processes and neural networks get more refined and advanced the end user benefits from these updates. Lastly, if you just don't have the time to deal with any of this yourself, the Audionamix team can extract your audio for you. It's not cheap at around $160 per hour, but they're experts at what they do and have an extensive list of credits with major film studios and record labels. If you need the best results and have the budget (but not the time), this might be a good option for you. I sent the Audionamix team a 30-second piece of audio for them to extract the vocals from and it was definitely the best sounding of the three extractions – although the entry-level XTRAX STEMS extraction came very close. To be honest, I like making music, not fixing it, so I hope I never need this software again, but it was interesting to learn more about it and I'm amazed at how far this technology has come. Additionally, if you do need to extract vocal stems, I can say the entry level XTRAX STEMS does a pretty amazing job of it at an affordable price.