These project ideas involve wind instruments. Following the project list is a set of notes about concatenative drum synthesis that may be helpful.
In the notes below, we discussed how multi-sampling -- systematically sampling a horn playing different notes, with different articulations and different volumes -- is the dominant current practice for horn parts played from a keyboard.
An alternative architecture for horn sampling would be to record an instrumentalist playing complete performances, and to then use this performance as a database that keyboardist "indexes into" in real time by playing a part. The hope is that more fluid parts will result, as the database contains natural transitions between notes. An implementation of this basic architecture is present in the Synful plug-in, as described here.
In this architecture, we are using the database for two reasons: to look up performance control data, and to look up waveforms. However, as we heard in the WX-7 examples linked in the notes, acoustic physical modeling synthesizers have existed for a decade which provide amazingly realistic horn sounds, given realistic performance control data (in our mp3s, from a WX-7).
The class project would be to begin with the concatenative real-time database architecture as used by products like Synful, but modify it so that the database is converted from an audio database to a control parameter database as an off-line operation. Ideally, new database would hold the sort of controller information a wind controller like the WX-7 generates. This parameter database could then be used for real-time lookups, to convert piano keyboard MIDI control to expressive WX-7 style control, which could then be used to drive a physical modeling synthesizer. This architecture has the advantage that the performance control database would be KBs, not GBs, and could easily fit in RAM.
Collaboratively working with a horn soloist or section in the studio is an invigorating experience for all concerned: the clock is ticking, and everyone in the room wants a successful take at the end of the day.
Contrast this experience with searching a loop library for the right horn section loop, or using a user interface like the one described for Liquid Saxophone. The robot "player" doesn't listen -- you do. You audition loops one by one, keeping in mind the role in your track the part needs to play, and try to stay focused.
Maybe there is a better way.
This project is about coming up with a user interface, and a plan for realizing the user interface using currently-known technology, that goes beyond a producer auditioning a lists of loops one by one.
The description for a project like this is necessarily vague: the goal is for some genuinely new and useful ideas to emerge from the class project. However, the following ideas should give a sense of a direction to get you started:
In an operational sense, one possible design would put the program and the user in a dialogue. Candidate concatenations of loops by the program would form sonic requests in the dialogue, and guidance from the user on how to improve the loop would form responses to the request.
[David -- your WX-7 timbre selection system goes here]
This lecture will focus on the wind instruments most commonly used in popular music: saxophone, trumpet, trombone. Following common practice, we use the term "horn" to refer to these instruments, although technically the saxophone is a woodwind and not a horn.
In pop music arrangements, horns play two major roles:
Soloists are usually hired to improvise a solo that fits the context of a song. The final track is usually a collaboration, to some degree, between the producer (who often has a "vision", in some abstract sense, for the solo) and the sax or trumpet player (who has the technique and the improvisation style to bring the producer's vision to life).
In horn section studio work, a wider variety of working styles are common. At one extreme, full charts may be written in advance for a section by an arranger. At the other extreme, a group of players who play as a section may be hired as team, with the understanding that they will develop appropriate lines for the song during the session in collaboration with a producer.
A good "novice guide" to the mechanics of writing section charts is available here. In addition, commentary on the horn section styles found in pop music may be found in the interview of Andy Bush in this article (page down to reach his interview).
A qualitative understanding of how trumpets and saxophones produce sound is essential for understanding the lecture. This article is an excellent tutorial on the simplified physics of wind instruments. That article, and this lecture, assumes a traditional playing style for the instrument, where the goal is to produce a pitched monophonic voice.
We summarize the article with a few key observations:
Taken as a whole, these characteristics let horn solo lines be expressive, in the way a vocal line is expressive (apart from the ability to convey words, of course).
Capturing the sound of horns in the studio is a challenging task: brass in particular has a large dynamic range (range of softest to loudest sound). This article is an excellent review of current practice for recording solo horns and horn sections. This set of engineer interviews is another good resource of engineering a horn session.
When analog keyboard synthesizers became popular in the late 1970s, hopes were high that realistic horn parts could be produced by the instrument: sawtooth waves were a good match for the harmonic series of an ideal blown pipe, and voltage-controlled low-pass filtering could model the dependency of brightness on loudness. Good descriptions on how to generate brass-like sounds from analog synthesizers can be found in these articles: a horn patch for a modular synthesizer and for a MiniMoog.
To listen to example analog brass sounds using these techniques, click on these links: sound 1, sound 2.
These sorts of sounds were used quite often in the 80s (and indeed, most modern keyboards include factory patches for "analog brass" emulation). However, at best, these sounds could be said to "evoke" a brass sound, much as the analog drum sounds we listened to last lecture evoke an acoustic drum sound (but would never be mistaken for the real thing).
What went wrong? Major issues to realistic analog brass synthesis at the time were:
In the next part of the lecture, we discuss research and commercial efforts to address each of these issue.
The importance of performance control to wind instruments was made clear to all in 1987, when Yamaha released the WX-7, a MIDI controller that emulated the user-interface of a saxophone. These photos show close-ups of the look of the WX-7, and this FAQ gives a good sense of the MIDI control information the WX-7 generates.
When the WX-7 was released, Yamaha's synthesizer product line was based on FM synthesis. Just like in analog synthesis, an FM horn patch played from a keyboard yielded an unconvincing emulation of an acoustic instrument. However, when FM patches were controlled by a WX-7 played by a talented saxophonists, the FM horn timbres came alive.
This mp3 of Sal Gallina playing a soprano sax FM patch on a WX-7 makes this point clearly. Even though the attack sound on the patch is not particularly authentic, the performance control makes the line sound like a convincing saxophone-like sound. This clarinet FM sound is also brought to life via WX-7 control.
Wind controllers are still available today, and are part of the arsenal of session players who specialize in doubling on many variants of reed instruments (example: Bob Magnuson, whose online WX-7 demo reel can be listened to web page).
In the mid-1990s, Yamaha developed a new synthesis engine for use with wind controllers. This monophonic synth attempts to model the physics of brass and woodwinds instruments. This MP3 gives a sense of the realism of its trumpet model, when controlled by a wind controller; this Sound on Sound review describes the basics of this physical modeling system.
Given this mp3, one would think the topic of today's lecture is now old hat: a ten-year old synthesis engine, when driven by a proper controller, is capable of producing quite realistic horn lines.
However, in a business sense, this product did not meet expectations. The Yamaha product line stays alive as a niche product, that caters to open-minded wind players. The basic problem is that the realism heard on these demos requires the use of a controller like the WX-7, and for a pianist, learning how to use a WX-7 well is almost as hard as learning how to play saxophone or trumpet well. Most composers have have had piano training, and thus are interested in using a keyboard instrument that could realistically emulate horns, but few composers are willing to put in the time to learn how to play a WX-7 well.
This market experiment was quite expensive for Yamaha -- many millions were spent on the investments of controller and physical model development, and although the technology worked well, it is doubtful they have broken even on the project.
However, a lesson was learned: new profit-making ventures in brass and reed synthesis focused on letting a piano keyboardist (and later, an operator of a DAW) replace horn players in some types of recording sessions. The lecture continues with examples of this sort of technology.
Horn section lines are easier for keyboard players to pull off than sax or trumpet solos, for several reasons:
Thus, instrument makers focused on improving the sound of horn section patches. As mentioned earlier in the lecture, one open issue was the unconvincing nature of the attack transient for brass: this is particularly important for pop section work, where a biting attack sound is important.
Roland addressed this issue with its D50 synthesizer, released in 1987. The D50 combined a digital emulation of an analog synthesis engine (oscillators, filters, VCAs) with a set of short samples to produce the onset of sounds known to be difficult for conventional synthesis (including horns). A full description of the D50 can be found here.
These mp3s ([patch 1] [patch 2]) are demos from the brass factory patches of the D50. These patches cross-fade a 100 millisecond sample of a brass onset with a traditional filtered-sawtooth analog synthesis patch. A key problem with this approach is fusion (i.e. tricking the ear into hearing the sequence of two sounds as one). The D50 solved the fusion problem by including on-board effects (reverb and chorous) that were used in the factory patches to mask the transition between sample and synth. Roger Dannenberg's group at CMU later solved the trumpet fusion problem without resorting to effects masking, by crafting synthesis waveforms that fused cleanly into the onset sound (see a paper on this work here, and listen to results here).
The D50 was an significant commercial success for Roland. Individual D50 factory presents became quickly overused, as artists began to use the patches on their recording projects (examples: pizzagogo soundtrack ).
The D50 shifted the market trend away from FM synthesis, and towards "workstation" instruments that used sampled waveforms in ROM as a key part of their architecture. Roland soon lost the lead in instruments of this type to Korg (Yamaha later also became a market player).
As ROM prices became cheaper (per bit) in each generation, these instruments (as well as the softsynth plug-ins for DAWs that compete with them) were able to improve the quality of their horn section emulation by "brute force":
The multi-sampling approach is now the state-of-the-art for playing horn parts from a keyboard. This method works well for horn section parts, once the keyboard player has become familiar with the setup of the sample library, as realism demands picking the right sample to match the section playing style.
There are several strategies for horn articulation control via the keyboard when using multi-samples. One approach is to use a footpedal to signal patch volume (and brightness), use key velocity to select an articulation style (staccato or legato), and use the performance wheels for vibrato depth (via the modulation wheel) and pitch-bend (via the pitch bend wheel). The Garriton sample libraries use this method, as described here.
Alternatively, if monophonic lines are assumed, pressing a new key while an old key is already depressed can be used to signal a legato transition to a new note. The Synful plug-in uses this method, and legato detection is also a part of the user interface of the Vienna Symphonic Library. Automatically detecting legato playing is not a new idea: it dates back to the analog-era Roland SH-101, if not earlier.
As we noted earlier in the lecture, horn solos and sections are often a collaboration between producers and instrumentalists, done without formal charts. For producers comfortable with this approach, scoring and playing horn section parts on a keyboard may not be practical: their talent lies in knowing what sounds right, not in performance.
In these producers, a better approach may be to buy "loop libraries" of horn sections playing a sets of riffs in different styles. An example of a horn section loop library is here. Most DAWs now have facilities for browsing loop libraries via keyword search, and for pitch-shifting and time-warping chosen loops to fit an existing track.
Sometimes, loop libraries are used to add horns to an existing track. In this case, a producer browses the library in search of the "right" loop(s). This can be a tedious job, and may be impossible if no library loop exists which is a good match to the track. Indeed, in certain styles of music, it is common practice to first choose loops that evoke the desired style, and then rewrite the composition to work well with the chosen loops.
The first products that extend the loop library concept from horn sections to horn solos are appearing on the market. The best example is Liquid Saxophone, whose product description (and audio demo) is available here. Liquid Saxophone is a plug-in that features a large set of jazz saxophone phrases, categorized into difference styles, than can be pasted together into a solo by the user. The plug-in uses resynthesis to allow limited editing of the phrases to fit a piece. An early review of the product is available here.
For some productions, loops and phrase sampling are no substitute for real players. The production may run into trouble if for logistical reasons finding in suitable horn players for a track is not possible (example: artists whose home base is far from major production centers).
To address this need, some session players now sell custom performances for a recording project over the Internet. For example, iHorns is run by a NY session musician, and will arrange and record a 5-piece horn section to match a supplied track for $195; a sax or flute solo is $85. Audio tracks are sent across the Internet in a way that ensures the solos will sync to the track. The main difficulty in this approach is communication: the spontaneous give-and-take of a producer and a soloist in the studio is replaced by emails and telephone conversations.