Giving DetroitBeat “Soul”

Posted on by

I decided to have another go at Yamaha audio style conversion. This time I converted the PSR-S950 “DetroitBeat” audio style to an all-MIDI style called “DetroitSoul”. The name “DetroitSoul” seemed appropriate since most of the MIDI rhythm parts are taken from the Yamaha preset style “Soul”.

The conversion process was a little different.

  • Load DetroitBeat to set up the mixing console and the DSP effects.
  • Enter Style Creator and create a new style. Turn off the audio part.
  • Copy all non-rhythm parts from the MAIN and FILL IN sections of DetroitBeat into the new style.
  • Identify candidate donor styles for the MIDI rhythm parts. The “Soul” and “MotorCity” styles sounded the best (e.g., similar groove, no conflicting beats, etc.)
  • Audition the candidate MAIN and FILL IN sections. “Soul” was the best fit.
  • Copy the RHY1 and RHY2 parts from “Soul” into the new style (DetroitSoul).
  • Build a table of possible donor styles for the INTRO and ENDING sections.
  • Audition donor INTRO and ENDING sections which have the same length as the INTRO and ENDING sections in DetroitBeat.
  • Copy the donor INTRO and ENDING sections into the new style.
  • Double check and fix up the variation effect (DSP1).
  • Copy the OTS settings for DetroitBeat into the new style.

The process was smooth, but still required a lot of button pressing!

Here is a table with INTRO and ENDING section lengths. This table could come in handy when converting other audio styles to MIDI.

Section   DetroitBeat  Soul  MotorCity  FranklySoul  70sChartSoul
-------   -----------  ----  ---------  -----------  ------------
INTRO 1        1        (1)      1           1            1
INTRO 2        4        (4)     (4)          5            4
INTRO 3        8         4      (8)          8            8
INTRO 4        1         1       1           1            1

ENDING 1       2         1      (2)          2            2
ENDING 2       5         3       3           2            3
ENDING 3       5         4      (5)          4            4
ENDING 4       1        (1)      1           1            1

The parenthesized items in the table indicate the source rhythm sections that were copied to the new style. Please note that nothing was copied into ENDING 2. I couldn’t find an appropriate replacement and the ending sounded good enough by itself. Sometimes the right thing to do as a musician is lay out!

ENDING 3 ran on a little too long and the extraneous beats sounded like the drummer made a mistake. I shortened the section length to 4 bars.

DSP1 is configured as a SYSTEM variation effect. Here are the effect settings and parameters:

Category: REAL DIST
Effect: ST AMP VT

COMP SW          ON
COMP SUSTAIN     0.4
COMP LEVEL       6.0
DIST TYPE        Crunch
DIST DRIVE       7.8
DIST EQ          Mid Boost
DIST TONE        7.0 
DIST PRESENCE    6.0
DIST OUTPUT      24

Overall, the end result is musical. The Soul beat is not quite the classic Motown groove. (Think “I Can’t Help Myself (Sugar Pie Honey Bunch” by the Four Tops.) I have a few phrases from the old Yamaha QY70 that might fit and someday I’ll look into it. In the meantime, enjoy the new style DetroitSoul.

60sSuperGroup mash-up

Posted on by

In my last post, I started the process of converting the PSR-S950 audio style “60sSuperGroup” to all MIDI. I like the 60sSuperGroup audio style, but the really strong back-beat in MAIN A makes it hard to use that section on anything other than the Beatles song “Ticket To Ride.” In order to make this style more generally useful, I’m replacing the audio rhythm tracks with appropriate MIDI tracks. Last time, I had worked on the MAIN and FILL IN sections and now it’s time to attack the INTROs and ENDINGs.

Here’s my process.

  • Find a style which is similar to 60sSuperGroup. 60sVintageRock is a good alternative because it is your basic Mark II rock and roll.
  • Load 60sVintageRock into Style Creator.
  • For each section, copy the non-rhythm parts from 60sSuperGroup into the new style.
  • Change the section lengths to match the source style 60sSuperGroup.
  • Save the new style as 60sFabFour.
  • Listen to each new section critically.
  • If a section doesn’t work musically, copy a section from a different candidate style.
  • Edit DSP effects to match 60sSuperGroup.
  • Save a bunch of intermediate copies along the way in case you need to back up to an earlier version.

Overall, the MAIN and FILL IN sections from 60sVintageRock were a good match and sounded pretty good. The one measure INTROs were OK, too. The longer INTROs and ENDINGs were more of a problem. So, I identified a few alternative candidate styles and built a table of INTRO/ENDING section lengths to find and try alternatives. Here’s the table:

           Target:
         60sFabFour  60sVintageRock  60sPopRock  VintageGtrPop
         ----------  --------------  ----------  -------------
INTRO 1      1             1              1            2
INTRO 2      2             4              5            4
INTRO 3      5             9              4            9
INTRO 4      1             1              1            1
ENDING 1     3             3              3            2
ENDING 2     4             4              3            3
ENDING 3     5             6              7            5
ENDING 4     1             1              1            1

I tried to use alternatives that were the same section lengths as 60sSuperGroup and the new style 60sFabFour.

ENDING 2 was the most difficult to nail. I tried different alternatives and then needed to shorten the section length to get rid of some beats that ran on — kind of like Ringo didn’t know when to stop. Here are the final source styles for the INTROs and ENDINGs: 

           Target:
         60sFabFour  Source style
         ----------  --------------
INTRO 1      1       60sVintageRock
INTRO 2      2       VintageGtrPop
INTRO 3      5       60sPopRock
INTRO 4      1       60sVintageRock
ENDING 1     3       60sVintageRock
ENDING 2     4       60sPopRock
ENDING 3     5       VintageGtrPop
ENDING 4     1       60sVintageRock

A lot of luck and trial and error is involved here. Luckily, the alternatives fit pretty well.

DSP1 is configured as a SYSTEM variation effect for both 60sSuperGroup and 60sVintageRock. However, the parameters are different. The guitars are sent to the SYSTEM effect in order to get a VOX AC30 amp “chime”. The final effect parameters are taken from 60sSuperGroup:

Category: REAL DIST
Effect: ST AMP VT

COMP SW          ON
COMP SUSTAIN     0.4
COMP LEVEL       6.0
DIST TYPE        Crunch
DIST DRIVE       7.8
DIST EQ          Mid Boost
DIST TONE        7.0 
DIST PRESENCE    6.0
DIST OUTPUT      24

This uses a Real Distortion effect and you will need to change this when porting the style to an older model keyboard. I did not change the OTS settings in the new style since I was happy with the OTS settings from 60sVintageRock.

I put a copy of the 60sFabFour style on the Music Gallery page.

S950 audio style mash-up

Posted on by

The Yamaha PSR-S950 and Tyros 5 arrangers provide audio styles as well as conventional, pure MIDI-based backing styles. Audio styles replace the MIDI-based rhythm tracks with an audio track of a (human) drummer playing a kit. The remainder of the backing track is provided by MIDI. On the up side, the audio rhythm parts have more nuance and sound pretty darned good. On the down side, the audio track in a style cannot be modified or eliminated nor can they be replaced by a user’s own audio track. Whether this limitation is a quirk of the Yamaha software or a permanent feature remains to be seen.

One of the S950 audio styles is “60sSuperGroup.” I’ve been off in Pepperland trying to pull some Beatles tunes together. I kept gravitating back to 60sSuperGroup for backing, but the drum back-beat in Main Section A is so strong that it doesn’t fit with hardly anything other than the song “Ticket To Ride.” It would be great to apply the non-rhythm parts to other songs.

Time to replace the audio rhythm parts in the 60sSuperGroup audio style. Unfortunately, one must work around the limitations of Yamaha’s software. The MIDI drum style “60sVintageRock” is roughly the same tempo and its rhythm parts are your basic Mark II rock and roll — in other words a good candidate style for a mash-up.

First, I loaded 60sSuperGroup, got into Style Creator and tried copying the 60sVintageRock rhythm parts into 60sSuperGroup. No joy. Once a style is an audio style, it’s always an audio style. Further, you cannot store the audio style to an external device like a USB jump drive. Yeah, the manual says this explicitly, but it was worth a try. No need to go down that rat-hole again.

So, here’s the process that I followed. I loaded 60sVintageRock as the base style and got into Style Creator. I then copied the non-rhythm parts from 60sSuperGroup into the new style which I called “60sHybrid”. I did this for MAIN A-D, FILL IN A-D, BREAK, INTRO 1 and ENDING 1. I’m not a big fan of long preplayed intros and endings, plus I didn’t know how well the longer intros and endings would mash up. Even without these additional intros and endings, this was more than enough button pushing for one day!

Here is a side-by-side comparison of INTRO and ENDING lengths:

         60sSuperGroup  60sVintageRock
         -------------  --------------
INTRO 1        1              1
INTRO 2        2              4
INTRO 3        5              9
INTRO 4        1              1
ENDING 1       3              3
ENDING 2       4              4
ENDING 3       5              6
ENDING 4       1              1

I’m not sure how to fix up these up as yet. Suggestions?

The two styles have different instrument-to-style part assignments. Here’s the instrument information for 60sSuperGroup:

Part  Vol  Pan  Var  Instrument         
----  ---  ---  ---  -----------------
RHY1   54  C      0  PowerKit1          OFF
RHY2   72  C      0  RealDrums          OFF
BASS   66  C      0  Mega VintageFlat   ON
CHD1   51  L28  127  Mega SingleCoil    ON
CHD2   42  R30  127  Mega SolidGuitar2  ON
PAD    48  C      0  Mega 12StringGtr   ON
PHR1   44  R32  127  Mega SingleCoil    OFF
PHR2   70  C      0  GrandPiano         OFF

The rhythm channels are both OFF because all drum/percussion is provided by the audio track. Here is the instrument information for 60sVintageRock:

Part  Vol  Pan  Var  Instrument         
----  ---  ---  ---  -----------------
RHY1   54  C      0  PopLatin           OFF
RHY2   72  C      0  RealDrums          ON
BASS   66  C      0  Mega VintagePick   ON
CHD1   51  L28  127  Mega SteelGuitar   ON
CHD2   42  R30  127  Mega SolidGuitar2  ON
PAD    48  C      0  CurvedBars         ON
PHR1   44  R32  127  Mega SingleCoil    OFF
PHR2   70  C      0  Harmonica          OFF

Since nothing was copied to INTRO 2-4 and ENDING 2-4, the instruments and sound for these sections do not match the sections copied from 60sSuperGroup. The mismatch is readily apparent when played.

DSP1 is configured as a SYSTEM variation effect. I needed to edit the effect parameters in order to get that VOX AC30 amp chime. Here are the parameters; they are the same as 60sSuperGroup DSP1:

Category: REAL DIST
Effect: ST AMP VT

COMP SW          ON
COMP SUSTAIN     0.4
COMP LEVEL       6.0
DIST TYPE        Crunch
DIST DRIVE       7.8
DIST EQ          Mid Boost
DIST TONE        7.0 
DIST PRESENCE    6.0
DIST OUTPUT      24

It’s been an interesting experiment so far! The resulting mash-up should be quite useful when tracking up-tempo, early Beatles rock and roll.

PSR/Tyros XG effects

Posted on by

I’ve been busy making my PSR-S950 gig-ready. I’ll describe the overall set-up in a separate post as soon as I have a little more time.

Part of the job involves converting some of my General MIDI 2 backing tracks to the PSR-S950 sound engine. The PSR (and Tyros) sound engine follows the Yamaha XG architecture and the sound engine responds to Yamaha XG System Exclusive (SysEx) MIDI messages. The XG SysEx messages configure tone generation and the effects that are applied to the tone generator outputs. Let’s limit the discussion to MIDI song sequencing and assume that there are 16 MIDI channels and each channel is routed to a separate tone generator. We’ll ignore style parts and how to tweeze tone generation in order to keep this discussion focused on effects.

There are two types of XG effects: system effects and insertion effects. We are already quite familiar with two common system effects: chorus and reverb. Potentially, the output from any tone generation channel can be sent independently to chorus and/or reverb. System effects routing and control follows a send-return model like a well-featured analog mixer. The amount of signal sent to the (virtual) chorus or reverb unit from a given channel is determined by a send level. Chorus and reverb are used so frequently that “standard” MIDI controller numbers are assigned for reverb (CC91) and chorus (CC93) depth.

Just like the real-world mixer model, an insertion effect is a channel specific effect. The output from a tone generator is sent directly to the input of the insertion effect and the output of the insertion effect becomes the output from the channel. An insertion effect belongs to one and only one channel (tone generator). The XG architecture does not allow insertion effects to be chained. So, if you need a chain of guitar-oriented effects, you need to read through the effects table in the Yamaha data list and find a multi-effect that does the job.

Now, life gets interesting. The XG architecture defines two kinds of (virtual) effect units: EFFECT1 and EFFECT2. Each kind of effect has certain capabilities and SysEx messages.

XG EFFECT1 has a special name: the Variation effect. The Variation effect is very (no pun!) flexible. It can function as either a system effect or as an insertion effect. In PSR/Tyros-land, the Variation effect is assigned to the DSP1 unit. That’s why the user and reference manuals have special rules and conditions that govern the use and configuration of DSP1. On certain specific PSR models such as the PSR-S750, DSP1 often supports more effect types (e.g., kinds of distortion or whatever) than its cousins, DSP2, DSP3, DSP4, etc. If a song uses DSP1, the set-up information in your MIDI SMF file, at a minimum, must send SysEx to choose system or insertion mode and the effect type.

XG EFFECT2 effects are optional, that is, an XG keyboard does not need to implement any EFFECT2 effects. (EFFECT1 is also optional, but such a keyboard is only “XG-Lite” compliant like the PSR-E443.) The old QY-70 sequencer, for example, implements EFFECT1, but doesn’t support any EFFECT2 effects. The Yamaha Mobile Music Sequencer only supports EFFECT1, too. In PSR/Tyros-land, the number of ancillary DSP units (DSP2, DSP3, DSP4, etc.) determine the number of supported XG EFFECT2 effects. All XG EFFECT2 effects are insertion effects. Period. No choice.

With this background in mind, you will need to take some time to study the XG effects-related parameters in the MIDI message section of the PSR/Tyros manual. This is time well-spent.

Let’s take a look at a few details, though, in order to see how all of this fits together. First, here are a few points to summarize EFFECT1, the Variation effect.

  • XG EFFECT1 is assigned to PSR/Tyros DSP1.
  • This effect unit provides reverb, chorus and other insertion/system effects.
  • The Variation effect is configured for either insertion or system mode.
  • The Variation effect is shown in the mixer console effect window.
  • As an insertion effect, the send level is 127 and cannot be changed through the mixer console window.

The following three SysEx messages configure the Variation effect as an insertion effect. All numbers are hexadecimal. The effect type is MSB:5F and LSB:20, which is MSB:95 (decimal) and LSB:32, known to human beings as Real Distortion “MLT DS SOLO.” It’s a guitar multi-effect suitable for a lead guitar solo.

     Start SysEx message
     |  Yamaha ID
     |  |  Device number
     |  |  |  Model ID
     |  |  |  |                 End SysEx message
     |  |  |  |                 |
    F0 43 10 4C 02 01 40 5F 20 F7      Choose effect type
    F0 43 10 4C 02 01 5A 00 F7         Variation connection (00:insertion)
    F0 43 10 4C 02 01 5B 00 F7         Variation part
                 |  |  |
                 EFFECT1 parameter addresses

These three messages are called “XG Parameter Change” messages because they change an XG control parameter (e.g., effect type) stored at a particular address (e.g., 02 01 40). Most of the message is reusable boilerplate like the Yamaha ID, device number and model ID. Since we are using the Variation effect in insertion mode, we must assign the effect to a part, also know as a MIDI channel. In this case, we are assigning the Variation effect to Part 1 (MIDI channel 0).

Here’s a few bullet points to summarize what we know about EFFECT2.

  • XG EFFECT2 is assigned to one of DSP2, DSP3, DSP4, …
  • XG EFFECT2 is always an insertion effect.
  • The DSP unit is selected via the XG parameter address in the SysEx message.
  • The insertion effect must be assigned to a (song) part/MIDI channel.

Here is a quick example of two SysEx messages to set up an insertion effect on part 2 (MIDI channel 1) on DSP4.

                 EFFECT2 parameter address
                 |  |  |
    F0 43 10 4C 03 02 00 4E 10 F7      Choose effect type (0x4E 0x10)
    F0 43 10 4C 03 02 0C 01 F7         Variation part
                    |     |
                    |     Part 2 (MIDI channel 1)
                    00:DSP2 01:DSP3 02:DSP4

The first SysEx message selects the effect type (XG parameter address: 03 02 00). The effect type is MSB:4E and LSB: 10 which is MSB:78 (decimal) and LSB:16 (decimal). This is AUTO WAH1. The second message assigns DSP4 (02) to MIDI channel 1. Please note that the DSP unit is selected by the second byte in the XG parameter address.

This information should be enough to get you started. From here, I recommend reading about the Yamaha XG tone generation and effects architecture.

Yamaha, at one time, published diagrams showing Tyros 2 and XG effect routing. Unfortunately, these helpful diagrams are now hard to find. Here are links to the Tyros 2 effect diagram and the MU-128 XG effect diagram.

Sparkfun Danger Shield

Posted on by

Sparkfun is one of my favorite companies. I like their products and their service is very good and reliable. I’ve learned a lot by studying their designs and I especially like their commitment to education.

Previously, I built two Sparkfun kits: the Redboard PTH Arduino and the MIDI break-out board. I have several more kits on hand to satisfy the occasional urge to solder!

Recently, I built a Sparkfun Danger Shield, hoping to use it as part of a MIDI drawbar controller. The code for the controller is still a work in progress. So, in the meantime, here is a micro-review of the Danger Shield.

The Danger Shield is a “jack of all trades” for input to an Arduino. It has three large sliders, three momentary contact buttons, one temperature sensor, one light sensor, and a capacitive touch sensor. The Danger Shield also provides basic output/display capability, too. The shield has two yellow LEDs, a seven-segment display, and a buzzer. For the MIDI controller project, I’m mainly interested in the sliders, buttons, LEDs and seven-segment display. However, I can see some creative possibilities for the other sensors in MIDI control.

Kit assembly went quite well, taking about two hours total. I think that a beginner could put one together without too much trouble. There was only one minor hang-up. The on-line assembly instructions are out-of-date. Two decoupling capacitors were added to the design after the instructions were published. One capacitor is mounted near the temperature sensor and the other capacitor is mounted near the shift register integrated circuit. This could trip up a beginner since they will have two small parts left over if they simply follow directions!

DangerShield

Sparkfun thoughtfully provide an Arduino Sketch (program) to test the sliders, buttons, etc. This is a great idea and I wish that more companies provided test programs with their products. When you build a kit, you really want to know if everything works before you design the kit into an experiment or prototype. Unfortunately, the test program expects the cap sense code to be installed as an IDE library. This dependency could trip up a beginner since they would need to learn how to install library code before running the test program. Since I didn’t intend to use the cap sense pad right away, I commented out the cap sense code and tested everything else.

The one thing that surprised me is the physical size of the shield. It is much bigger than the standard Arduino footprint. I had originally intended to stack the Danger Shield, the MIDI break-out board and the Arduino on a Liquidware side-by-side extender. This approach would have saved me the effort of whipping up a 5-pin MIDI OUT port. Unfortunately, the large size of the Danger Shield prevents much stacking.

I decided to prototype on an Arduino UNO which is installed on a plastic Arduino and breadboard holder. The Danger Shield is stacked on top of the Arduino UNO. The 5-pin MIDI OUT port resides on the breadboard. I built a 5-pin DIN break-out board to securely attach the 5-pin connector to the broadboard as well as provide a way to make necessary connections to the Arduino +5V, ground and TX pins. This quick-and-dirty break-out board should aid future experiments, too, and is a good investment of time. Finally, I wrote a quick test program to drive MIDI data through the output port and to make sure that it was electrically sound before I connect it to an expensive synthesizer or arranger workstation.

All in all, I recommend the Danger Shield. It’s possible to build a pretty decent user interface given a little bit of creative thought. The sliders and buttons are robust and should endure much abuse during testing.

MMS as a tool

Posted on by

I’ve got a lot more experience with the Yamaha Mobile Music Sequencer (MMS), so it’s worth passing along an update. I’ve been composing backing tracks for classic soul jazz tunes. If you would like to hear the results, please browse over to the Music Gallery.

MMS is a phrase- and section- oriented composition tool. A song is a sequence of one or more sections and a section is a group of phrases that play together. The phrases follow the section’s chord progression. Thus, it’s easy to pull a section together given a chord progression from a lead sheet and a library of drum, bass, guitar and piano phrases (MIDI loops).

This approach works great for a simple tune like “Memphis Underground.” Memphis Underground is built on a single chord (C7) and simple bass figure that repeats ad infinitum. Just set the chord progression for each section to C7 and stack drum, bass, guitar and electric piano phrases. Vary the arrangement by stacking different phrases in each section and lay down the different sections in the song. I recorded the simple flute part that makes up the head (the main melodic theme) into a phrase of its own. Finally, I recorded solo parts into the MMS song screen because it didn’t make sense to split the solos into separate phrases. Overall, this approach worked out pretty well.

Life got a little more interesting with “Watermelon Man” and “Comin’ Home Baby.” These tunes are 16- and 12-bar blues. It would be great to arrange the songs from short 4-bar phrases and just let the phrases follow the chord progression. However, when fills are placed at the end of a 4-bar phrase, the fills do not always play at the most musically appropriate points in the tune! I resolved this problem by increasing the phrase length to 8 bars. I also recorded the head into a phrase of its own. It’s handy to play back the head while stacking phrases even if you intend to record the head along with the solos in the MMS song tracks. Melodic phrases such as these must be set to by-pass transposition.

Then there are tunes like “Tough Talk” and “Put It Where You Want It.” These tunes are based on one or more musical hooks that are essential to the character of the song. Generic bass or piano phrases just don’t cut it. I had to record phrases to cover the hooks and the head. Now, on-the-fly transposition guided by the section chord progression really starts to fight you! I wound up recording full, chorus-long phrases (all 12-bars), effectively ignoring (defeating) chord transposition by the sequencer. Each section has only one chord (e.g., F7 or C7) which simply determines the key for the tune. The hook phrases must be set for “parallel” transposition.

Put It Where You Want It is a work in progress. This tune is even more complicated to sequence because it has three major sections, each with a distinctive hook and theme. Stay tuned!

Overall workflow with MMS has been good. The Mixdown feature makes it easy to create a WAV file. MIDI (SMF) export is also easy. I’m using iTunes File Sharing to move the WAV and MID files to a PC where I convert the WAV to MP3 and add General MIDI (GM) SysEx and voices to the SMF file. iTunes File Sharing is a lot less hassle than I originally anticipated. The latest version (3.1) of MMS adds Dropbox, but I haven’t updated as yet.

There is one minor recurring problem. The MMS tone generator is a subset of the Yamaha XG standard and includes extended XG drum kits. This is good for musicians who are working on XG-compatible and/or Yamaha instruments because the GM drum kit is quite limited. However, the extended MIDI notes outside of the GM range do not map to the same percussion sounds on non-XG equipment, e.g., Roland Sound Canvas. So, I have had to edit the MIDI file and remap notes to make the drum parts truly General MIDI compatible.

Workflow is essentially in one direction only. I think the software developers see MMS as a mobile sketchpad where a musician jots down ideas that are transferred to and finished on a computer-based DAW. MMS cannot import results from the DAW. So, once you start editing with your DAW (e.g., SONAR or Cubase), you’re committed.

Well, there you have it. The true worth and limitations of a software tool like MMS are only apparent when taking on complicated, real-world problems. I’m still enthusiastic about MMS, but I’m also more knowledgeable and wary of its limitations. The song/section/phrase structure can definitely fight back at times!

Tyros 4 teardown

Posted on by

I stumbled across this Yamaha Tyros 4 teardown on Youtube. Although the video is a little bit long-winded, you get a peek under the hood of the Tyros 4 and see a few of the circuit boards.

Unfortunately, the video is short on the kinds of details that you would find in a Yamaha service manual. I wish that the producers had taken a close pan of the main logic board with enough resolution to read the part numbers. The narrator read off the part number of the main CPU, Renesas R8A77310D, but then stated that he couldn’t find information on the web. OK, here’s the info. It’s a Renesas SH7731 processor containing an SH4AL-DSP core. The maximum clock speed of the SH7331 is 333.4MHz. The SH-4 core is a rather powerful, DSP-capable core.

The narrator also mentioned an Altera Cyclone. This is a field programmable gate array (FPGA) which probably provides some glue logic like a bus bridge. Since it’s an FPGA, it could be programmed to do just about anything.

The video shows at least two other very large scale integrated circuits. Two of these are probably SWP51 tone generators. The SWP51 has two 16-bit wave memory ports (each with separate address and data ports). Therefore, the IC package is big and has a lot of pins. The SWP51 is where Yamaha keeps the secret sauce. They have never published papers about it and with good reason; Korg, Roland and Casio would probably love to know what’s inside, too!

Given that the Tyros 4 has Vocal Harmony 2, it most likely has an SSP2 chip.

Leaving the video aside, here’s a few “big picture” thoughts.

First, it’s interesting to see how Yamaha have been riding the CPU and memory technology curves. They have used successively more powerful SH architecture CPUs over the years. They clearly have deep knowledge and competence with this architecture. Memory-wise, they have progressed from mask ROM to bulk programmable ROM (P2ROM) to NOR/NAND flash. NOR/NAND flash is so widely used in the PSR-S950, for example, that the entire machine — in theory — could be reprogrammed.

Next, if I were Yamaha, I would consider using an ARM system on a chip (SOC) in the entry level and possibly the lower mid-range keyboards. They could achieve a higher level of functional integration with ARM and still obtain low power consumption. Further, they could provide superior tone generation at the entry level. The proof point is the Yamaha Mobile Music Sequencer (MMS) app on the Apple iPad. MMS on ARM supports eight tracks of playback with a polyphonic, XG-like software tone generator. MMS provides basic XG variation effects which are not current entry-level features. Of course, this means that Yamaha is willing to leave its comfort zone with the SH architecture!

Yamaha arranger product family

Posted on by

Even I found the last post to be down in the weeds and confusing. So, here’s a better summary.

I went on an Internet dumpster dive over the weekend to find more service manuals for Yamaha keyboards. I’ve always been interested in the architecture of product families and curiosity got the best of me, again.

The Yamaha arranger keyboard family has four main tiers: 1. Premium, 2. Upper mid-range, 3. Lower mid-range and 4. Entry. These are my names. Each tier has its own hardware architecture. Let’s take it from the top.

Keyboard Main CPU Sub CPU Tone gen Wave ROM
Tyros 2 128MHz SH7727 200MHz SH7709S 2xSWP50 6x256Mbit
Tyros 3 128MHz SH7727 200MHz SH7206 2xSWP51B 4x512Mbit
Motif XS 400MHz TX4939 None 2xSWP51 2x512Mbit

Tyros keyboards form the premium tier. I could find service manuals for only the Tyros 2 and Tyros 3; I’ll bet that the Tyros 4 has a similar internal architecture. The Tyros has two CPUs: a main CPU to handle the user interface and peripherals (e.g., the hard disk) and a sub CPU for digital signal processing (DSP) and to manage tone generation. The main CPU bus and the sub CPU bus are connected through a bridge (implemented in gate array technology). The main and sub CPU each have their own SDRAM and program ROM. (This is true for the other tiers as well and I won’t mention it again.)

Tone generation is performed by two SWP5x integrated circuits. The SWP51 is the current series. The wave ROM is split into an upper and lower bank and is shared by the two SWP5x tone generators on a common wave memory bus. SWP51 wave ROM is dedicated to tone generation. The SWP51 implements higher waveform sample compression. Thus, the Tyros 3 actually has less physical wave ROM than the Tyros 2, even though the Tyros 3 implements memory hungry Super Articulation 2 (SA2) voices. Interestingly, there is an audio backchannel from one of the SWP51s to the hard disk subsystem. The audio data passes through an “audio transformer” circuit on its way between the SWP51 and the hard disk subsystem.

BTW, I resist making any guesses about the inner design of the new Tyros 5.

I included the Motif XS in the table for comparison. The Motif’s TX4939 is a MIPS architecture processor while the Tyros CPUs are SH-2/SH-3 architecture. (Completely incompatible, of course.) The Motif XS runs Monta Vista Linux. The tone generation architecture is very similar to the Tyros: two SWP51s and two banks of wave ROM on a shared bus.

Keyboard Main CPU Sub CPU Tone gen Wave ROM
PSR-S750 135MHz SWX08 None 1xSWP51L 2x1Gbit
PSR-S950 256MHz SH7331 SSP2 1xSWP51L 2x1Gbit
PSR-S710 128MHz SH7727 None 1xSWP51L 2x256Mbit
PSR-S910 128MHz SH7727 135MHz SWX02 1xSWP51L 2x512Mbit
TMS320DA150

The upper mid-range keyboards have a main CPU. However, only the S9xx keyboards have auxilliary processors. The older S910 has a sub CPU (SWX02) and bridge, an arrangement which is similar to Tyros. The S910’s TMS320 (Texas Instruments DSP) implements the MP3 CODEC. The S950 does not have a sub CPU. Its main CPU, however, has a much higher clock rate and probably took over the workload performed by the S910’s sub CPU and MP3 CODEC. The S950’s SSP2 processor is a Yamaha custom IC for vocal harmony processing.

Upper mid-range keyboards use one SWP51L integrated circuit for tone generation. Wave ROM is split into two banks and is connected directly to the SWP51L. Current generation keyboards (S750 and S950) make wide use of flash memory. More than ever, product features are determined by code and content alone.

Keyboard Main CPU Sub CPU Tone gen Wave ROM
PSR-S650 135MHz SWX02 None On SWX02 1x512Mbit
PSR-S550 135MHz SWX02 None On SWX02 1x256Mbit

The lower mid-range keyboards have only a main CPU: an SWX02. The SWX02 has a Renesas part number and is probably an SH architecture machine. Tone generation is integrated into the SWX02. The SWX02 has a dedicated interface to the wave ROM. The S650 has only 25% of the S750’s physical wave ROM.

Keyboard Main CPU Sub CPU Tone gen Prog/Wave ROM
PSR-E213 128MHz SWL01 None CPU 1x32Mbit
PSR-VN300 128MHz SWL01 None CPU 1x64Mbit
PSR-E403 128MHz SWL01 None CPU 1x64Mbit

The entry level keyboards are, of course, the most cost and price sensitive. The entry level keyboards can also run on battery power thereby imposing further power and performance limitations. Entry level keyboards do not provide much in the way of voice-level effects. All three of the example keyboards have a single SWL01 CPU and are all clocked at 128MHz. The ROM contains both the program and waveform data. Thus, waveform data is read across the CPU memory bus along with all of the usual program/data traffic. Physical ROM size is very much smaller then even the S550/S650. The component count is very low and the circuit boards are quite simple.

A follow-up on the Yamaha SWP51

Posted on by

Here’s a follow-up on the Yamaha SWP51 tone generator.

Sound On Sound (SOS) mentioned that the SWP51 tone generator was first used in the Yamaha Motif XS family. The Motif XS has two SWP51 ICs with a master/slave relationship. Each SWP51 has 8MBytes of dedicated DSP SDRAM. The two SWP51s share wave ROM arranged in two banks (high and low) of 512Mbits each for a total physical capacity of 128MBytes. Each wave ROM device is a Spansion S29GL512N10TFI020 which is a 32M by 16-bit parallel NOR flash memory. (Yay, Spansion. I ate lunch in their cafeteria in Austin.)

SOS and others claim that the SWP51 performs sample compression. The Yamaha specifications state wave ROM capacity at “355MBytes when converted to 16-bit linear format,” meaning uncompressed size. The waveforms are compressed in order to fit into 128MBytes of physical memory.

There is only one thing that we can conclude for sure. The PSR-S750 and PSR-950 have twice the physical wave ROM space as the Motif XS and MOX (256MBytes vs. 128MBytes).

The Motif XS with two SWP51s has 128 voice polyphony while the MOX has 64 voice polyphony. Thus, the MOX most likely has only one SWP51. The PSRs have 128 voice polyphony. If the later version of the SWP (SWP51L) has the same number of tone generating and DSP elements as the first version, then the arranger keyboards are deploying the elements differently than the Motif family instruments. Without knowing the internals of the SWP51 and its variants, this is pure speculation.

By the way, the main processor in the Motif XS is the Toshiba TX4939 RISC CPU. The CPU is clocked at 400MHz. The TX4939 is a MIPS architecture processor with several integrated I/O controllers and interfaces. The TX4939 is capable of generating an audio sampling clock and supports PCM input/output. The processor runs Monta Vista Linux. Yamaha has an equity stake in Monta Vista and distributes the GPL’ed source code three years after initial product release. Yamaha have not released source for the arranger keyboards, so most probably, the arrangers use a different embedded operating system.

The MIPS instruction set and the Renesas SH-4 instruction sets are not compatible. If the arranger and workstation product lines share software-level functionality, it must be at the source code level.

Just in case you aren’t confused enough already. The Tyros 3 has two SWP51B ICs and two SH-series CPUs (Renesas SH7727 SH-3-DSP main, and SH7206 SH-2A sub). The sub CPU handles the SWP51Bs while the main handles the user interface, etc. The microphone input is routed to one of the SWP51B ICs. Playback/record audio is routed from the other SWP51B to a gate array that interfaces to the hard drive subsystem. Wave ROM consists of four 512MBit mask ROM devices arranged in two banks for 256MBytes of total physical ROM. Tyros 3, by the way, was the first Tyros with Super Articulation 2 (SA2) voices.

What’s inside of a Yamaha arranger?

Posted on by

Curiosity finally got the better of me and I decided to find out what’s inside of the Yamaha PSR-S750/S950 arranger keyboards. Fortunately, Yamaha provides service manuals for its products. The manuals have block diagrams and schematics as well as disassembly information, etc.

My first impression is that the S750 and S950 are quite different beasts inside even though a fair amount of user-level functionality is similar between the two products. However, some internal differences are pretty obvious and expected due to different features:

  • The S950 has a bigger set of voices and styles.
  • The S950 supports a wider range of effects on all four DSPs.
  • The S950 adds vocal harmony.
  • The S950 has a color display and can display lyrics and so forth through a video output.

Both products are relatively complex, multiprocessor systems, so the analysis below is greatly simplified.

As you might expect, both products have a main processor (CPU) to handle the user interface, the USB interface, and so forth. The S750 has an SWX08 CPU, which is most likely a Yamaha sourced SH3 or SH4 system-on-a-chip (SOC). The SWX08 has a Yamaha part code and is probably manufactured by Yamaha itself. The S950 has a Renesas SH7331 processor, which has an SH4AL-DSP CPU core. Yamaha has employed Hitachi/Renesas SH processors for many years. The SH4 is a reduced instruction set computer (RISC) that handles both general purpose computing and digital signal processing (DSP). The SH4AL-DSP can perform a multiply/add step in one clock cycle. Both machines are capable of handling some DSP duties on the main CPU. The SH7331 is clocked at 256MHz while the SWX08 is clocked at 135MHz.

The S750 program memory is 256Mbits. The S950 program memory is split between a 64MBit flash boot memory and a 4Gbit main program memory (Hynex HY27UF084G2M). The Hynex memory is 8-bit serial (512M x 8-bit) NAND flash memory. The address and data are clocked sequentially through an 8-bit port. Since this is a relatively low bandwidth interface, the program is loaded into SDRAM working memory first and then executed from there. The S950 working memory consists of four 128Mbit devices plus one 256 Mbit device for a total of 96MBytes ((4 * 16MByte) + 32MByte). I wouldn’t be surprised to find audio track data stored in the big NAND flash along with the program image. The S750 working memory is 64MBytes (2 * 256Mbit) of SDRAM.

Tone generation on both machines is performed by an SWP51L integrated circuit (IC). This is a custom Yamaha IC. The SWP51L has a 64Mbit by 16 bit SDRAM for DSP through a dedicated channel. The SWP51L is fed by wave ROM divided into HIGH and LOW banks. Each bank sends a 16-bit data stream to the SWP51L. Surprisingly, the wave ROM capacities are the same. The S750 and S950 have two banks of 1 Gbit NOR flash memory each (256MByte total).

Neither processor has a separate dedicated memory for downloadable expansion packs. The main CPU very likely reserves 64MBytes in the wave ROM for expansion pack samples. (“ROM” is a bit of a misnomer in this context.) Thus, one could expect to see larger expansion memory in future products when more wave memory is added.

The vocal harmony and display processing are handled by separate dedicated processors. The vocal harmony processor (SSP2) is connected to the output of the microphone analog-to-digital converter (ADC). SSP2 has its own dedicated DSP RAM and program memory. Each product has a display controller: the S1D13700 Embedded Memory Graphics LCD Controller on the S750 (black and white LCD) and the Yamaha Advanced Video Display Processor 7 (AVDP7) on the S950 (color LCD).

It’s interesting to look back at earlier Yamaha keyboards. The PSR-1500 and PSR-3000 were released in 2004. Here’s a table comparing past (2004) with present (2012).

PSR-3000 PSR-1500 PSR-S950 PSR-S750
SA 0 0 62 38
MegaVoice 10 0 23 18
Regular 261 273 571 523
Sweet 14 8 27 24
Cool 18 5 64 46
Live 19 1 39 29
Wave ROM 64MB 16MB 256MB 256MB

The Yamaha MOX and MOXF, for comparison, have 355MByte and 741MByte wave memory, respectively, when converted to 16-bit linear format.

The Super Articulation (SA), MegaVoice and Live voices are the most memory hungry. Both SA and MegaVoice voices need multiple articulations (multiple waveforms). The Live voices are sampled in stereo and require twice as much space as the equivalent mono (regular) voice. Of course, there are many other factors such as the number of multi-samples and loop length that affect memory usage and sound quality, so a grain of salt is needed when interpreting these numbers.