Guitar | Bass | Keyboard | Microphones | Mixers | Audio Interfaces | Monitors | Sequencers | Soft Synths | Live Sound | Drums | Club | Accessories | Blowouts
Understanding the Vast Matrix of MIDI and Audio data
Strategies for Converting MIDI to Audio from the tracking stage to Pre-mix stage
by Rich the TweakMeister
Has anyone seen the movie The Matrix? Suddenly the hero, Neo, realizes he's not in the world he thought he was. Reality is an illusion and the physical structures around him are made of "code" or data. I invite you to consider the sequencer in the same way. The Virtual Studio is an illusion. It only looks like a multi-track recorder and automated mixing console, filled with effects, compressors, eq, vintage synths, etc. That is the metaphor the application is using to help you understand it. What today's sequencer really is: it's a massive multi-threaded clock-based data processing system with a sometimes colorful graphical overlay of objects referencing classes and subsets of data. Imagine in your mind you can strip away the overlay of logic or cubase or pro tools and just look at the raw data processes going on "under the hood".
What would we see? We'd see MIDI data streaming into the system, being stamped with time, and directed to memory and to subroutines that generate further massive streams of audio numbers. It would join hundreds of other number crunching routines all working at the same time for the other "tracks" in the sequencer. Now imagine 32 tracks of this activity: MIDI going through huge software models of old synths and vintage effects; drum samplers spewing out data that lands exactly on the beat; convolution reverbs taking audio data, creating a second "wet" stream and running it through calculations that allow it to emulate famous concert halls and stage venues around the globe. Data then is diverted to busses and sorted through sends and returns of the virtual console, auxiliary busses start pouring out their data and it all gets "summed" to the final left and right data streams then out the door to the digital to analog converter. Amazing!
If we actually saw all this numerical and mathematical activity going on we'd see more screens of numbers racing around than our eyes could follow. Imagine it now. Hold the thought, Neo! (We are coming back to this point later.) The CPU, at the center of it all, is like an engine running flat out against the redline of the tachometer. Now if you were Neo, and you had to mold and interweave all these data pathways into two pathways (i.e., stereo) for a natural sounding song, how would you do it?
The point is a matter of Respect. Once you respect the processes going on "under the hood" you can then make intelligent strategies for making MIDI and audio work for you as you go through the stages of making a song. If you don't respect the underlying data activity you are creating you will end up fighting your DAW, who, in the end will shut you down and maybe even crash if you push too hard.
Positively put, respect the data processing and your DAW feels like it is working with you, you can harness its powers to accomplish tasks more effectively.
As we left off in the MIDI Basics class the recording and retrieval of MIDI data is at the heart of the modern sequencer. Recall that the major sequencers, Cubase, Logic, Sonar, Digital Performer began as MIDI only applications. Audio came along when computers were sufficiently fast enough to record audio in real time. Then software synths and samplers came about which took the MIDI data and generated audio on the fly, a process that puts incredible demands on the CPU.
Strategies for rendering and bouncing MIDI tracks to audio tracks
Without doubt, one of the most CPU intensive operations is to instantiate and play data through a software synth, a plugin which has its own knobs and sliders, that modifies the data and sounds in real time. The more soft synths you invoke, and force to play in real time, the greater the drain on your CPU. The weaker your computer, the fewer of these processes you can have. The other culprits are convolution reverbs, guitar amp modelers, vocoders, thickly sampled instruments from professional libraries.
Here are some ways you can get more out of your CPU.
1. The "Old School Method" Here, you use hardware MIDI synths and Samplers. When you use a hardware synth, the sequencer only has to record the MIDI data and send it back to the instrument on time. MIDI data is so small and efficient it does not put any strain on the CPU, even when you have scores of tracks. Remember the old Atari ST and Commodore 64 were able to do complete multi-track midi compositions. Even if you have a poor performing computer you can make this "old school" MIDI method work for you.
There are plenty of used MIDI synths and samplers on the market. Get a hardware mixer. Record audio tracks in the sequencer if you want, when they get back to the hardware mixer send them out to hardware processors to touch them up a bit or just leave them alone and we'll process later.
Once you have the song almost done you can start recording each of the MIDI tracks as an audio track. There is a bit of an art here too. If you have 48 MIDI tracks (and you might in some forms of orchestral electronica) are you going to want to sit there and record 48 separate audio tracks? I would not. Instead, arrange them logically on the screen and record in groups.
For example, your song might have a 4-piece brass ensemble. Mute all the tracks in the sequencer and play only the brass, recording the audio output of your synths and samplers to a single audio track. Do the same for other groups, such as strings, percussion, pads, etc. You would give bass its own audio track. If your drums are coming from your synths or samplers you may want to have a separate kick track and snare track and a stereo track with everything else. Again this is achieved by muting the tracks you don't want recorded in the pass. (or you can solo those you do want to record if you mixer has a "solo in place" function). In the end, you might have 6-9 audio tracks from the MIDI process. You do your mix on these tracks along with the vocal tracks and other tracks you recorded as audio from real instruments.
2. The "Render as you Go" Method. Essentially you use the "track freeze" function of your sequencer as soon as you can. Track freeze creates a temporary audio file in place of the real time midi to audio conversion. In the temporary file the output of the soft synth/sampler is combined with the output of the plugin processors on that channel strip. By freezing the channel you also disable all the realtime processes--that is what saves the CPU cycles. I
If your sequencer does not have this function you can do it manually by bouncing the track to audio, muting the MIDI track and bypassing the plugin softsynth and all processors on that channel. This completely frees the process from the CPU's activity, yet retains it so you can edit it later if you need to.
If you use Render as you go you will be freed to explore the full depth of your soft synths using multiple pass of controller tweaks. Because all the other processes are rendered down to audio, the CPU has the headroom it needs to tweak and automate the virtual synth without falling behind or halting the song due to an overload.
If you use bounce to audio you can group several channels together, premixing them with effects. Just as we did in the old school approach above, you can group tracks in the virtual mixer and bounce only those. But this time we never leave the digital domain. Freezing is good when you are tracking real fast and need to maximize your CPU power. Bouncing is for when you start preparing for the final mix. That is the advantage of bouncing over track freezing. The advantage of grouping tracks and committing to audio is to make the mix easier on your mind. It is much easier to mix 16 audio tracks than it is to mix 32 software instrument tracks and 10 audio tracks.
Ok lets talk a bit more about how to set up groups of tracks for bouncing.
Creating Mix Groups for Bouncing
Every mixologist is going to have their own biases towards grouping things certain ways, and all will tell you that it depends on the song and what each instrument needs to accomplish. Still we can put some rough guidelines down to get you thinking. Some I have already mentioned.
Lets say you have a song with the following tracks. I will highlight those in bold if they are extremely data intensive. Then I'll show you a hypothetical way to group them that will help us.
OK now. Please don't take those groups as "the" prescribed way to start a mix. Don't you dare show up on the forums and say "Tweak sez you have to bounce the hi hats and tambourine together". In many songs that's a bad idea. Remember, every mix will call for a different set of groups. You don't have to use groups or bounce. If you only had an 5 track song, it would be kind of silly to do so. Those with strong DAWS may tend not to even consider it with larger projects.
But I want you to see how we have 1) Made our coming automated mix much much simpler 2) probably cut your CPU usage to 15% of what it was before. This will give us the headroom we need for convolution reverbs, touch up compressors and eq, and other effects we may need. The lead vocal is still untouched and can be tweaked up to the end of the mix. The lead guitar solo, bass, kick and snare can also be re-tweaked all the way till the moment before the final mix.
Note also that even though you rendered many tracks down to a few, you should never throw away your source tracks or plugin settings. Just mute the track and put the softsynths, samplers and plugin effects in bypass mode. You might need to come back and do the percussion again if your first premix did not work.
Note how many variables are cut down to few. Imagine Neo's vision of all the simultaneous data processes under the hood. Still a lot going on, but the CPU is keeping up. More importantly, consider your own mental process and how much relief you will have only processing, leveling and panning 12 tracks rather than 26. If there was one tip underlying them all its to "think ahead" before you get too far into tracking your song. If you have respect for the data processing going on inside the DAW, you may be able to engage it as an ally in the process of song construction, rather than fighting it all the way. Remember, in the Matrix, Neo eventually discovered, through much trial and error, that he could manipulate the data processes around him. You can too. In your DAW, you are "The One".
Go to the Next Class
Go to the Previous Class
sets up ladders,
" Hilda Doolittle (1886–1961), U.S. poet.