Using DeMix to Separate STEMS скачать в хорошем качестве

Using DeMix to Separate STEMS

The STEMS in this video can be heard here:    • DeMix | SaveMe | Vocal STEM   Additional Notes! • General When scanning through the separations you are looking for shapes and their corresponding frequency range and the intensity of energy. If you see vertical lines, that's something with a fast attack. You can hone into that sound by familiarity with the frequencies of instruments. The you can utilize specified selection tools to try to capture only the necessary part of the visual. From that precise section you can fine tune the mixing of that separation to fill up a section of an alternate separation. Once the layer is merged appropriately, you can go to your target layer and further shape things to clean up your edits in the final version of the stem. • Backing For instrumentals, the post edits that take place will often involve digging into the backing section. Shapes, will stand out, or you can check the melody view to look for unwanted melodic bleed. You may want to port very low or very high frequencies to create a bonus Sound Design track as there may be ample additional elements and pieces that end up there and would benefit from having complemental material. Instead of just cutting a selection from the spectrogram, you can utilize the various threshold, frequency, or masking tools for more precise mergers between layers. • Percussion You'll generally want to touch up the highs in percussion, the AI seems to do a fair job of getting 90% of the significant percussive material in DeMix. Looking at the color will help you descearn what might be percussion that ended up in a backing track because it will have a similar color range because of the level. Cymbals tend to end up in backings for instance, parts of them any way, in those cases it may be a good idea to check that section of the backing with the masked eraser tool set to the appropriate mask level. In the video you can see a slight gap in the upper frequencies, that is an area I would potentially look to fill with content from another layer if needed to make the drums more natural (like the mix). • Bass For Bass you can get a more organic sound by exploring the lower regions in the backing. It's also worth exploring the percussive and backing laters for upper frequencies of bass parts which can be sliced out potentially. If you want to see what just bass actually looks like you can put just a bass recording inside the program and extract that. From there you can see what spectrum visualization that style of bass actually looks like and keep that in mind for the contexts of searching for similar shapes hidden within layers. Bass is another area where Sound Design likes to travel. Bass is generally easier to manage for a basic extract but a bit more of a challenge if you want a more convincing isolation. • Vocals Not covered in the video. The simpler the vocal the better the extraction. Especially with narrow vocal recordings. Fragments of vocals often end up in the backing track and are much more glaring when they end up in percussive separations. See the melody tool for a deeper dive into precision vocal or lead extraction. One common issue is that blend with cymbals because of sibilance. That is a case where making a more narrow cut in the similar frequency range will be more applicable. The eraser tool with masking with be of utmost use for a lot of the vocal fragmentation issues. Melody View * The melody view is a world unto itself, it was not covered in this video. It's essentially used to draw melodic lines over your track to help the AI take special notice of that content. It would be helpful in cases where there are vocals buried in effects that the AI does not recognize as vocals. You can also use the melody tool to help define lead instruments for removal. Tricks: • Remix the track prior to import to focus on key areas (meaning lower the volume of something in the source separation software, then reimport it). • Spectral Editors that work with layers work more like graphics editors. Spectral Layers even has a CLONE tool! • The oder of separation also can make a difference in the extraction/separation. • Different software will be using very different techniques, some will be more applicable to certain kinds of material than others. • If there are only certain instruments in one part of the track and other instruments in the rest, you can split this layer to make 2 stems. • You can keep separating on layers that have already been separated to try to pull more sound to match a specific layer. • Try to capture as many decent layers as possible prior to deeper editing, it makes the noise artifact removal run smoother. Check out more about DeMix here: https://www.audiosourcere.com #DeMix #AudioSourceRE #SourceSeparation #AudioSeparation #MusicSeparation #VocalRemoval #SpectralLayers #Xtrax #StemExtraction #audioSourceSeparation