Creating the database for MIRAI at UNCC

Creating the database for MIRAI at UNCC.

Following is a detailed description showing of the process of constructing sound objects. To illustrate the procedure we use the MUM Volume 6: Latin Grooves #1: Solo Instruments CD.

STEP 01: . Place the MUM's CD into the Drive and view the contents: As one can see, there are 79 "wav" tracks on this CD. Automation of this step is not an option. Track 01, not seen in the picture is named in the catalogue as:

01--08 marimba: single stroke and crescendo roll (2 entries per index) clearly, this is unacceptable for the MIRAI database as each folder will contain the grouping of only one set of sound implementation possible from an instrument. For example, in the violin sounds, we have separated, bowed, natural harmonics, muted vibrato, martele, pizzicato and bowed vibrato because the aforementioned form distinct classifications of a violins potential scale.

Herein, 01--08 marimba: single stroke and crescendo roll (2 entries per index) needs to be divided into marimba: single stroke and marimba: crescendo roll (2 entries per index) later in Step 3.

Automation of this Step is moot as only a human can determine when the MUMS, or any other dtabase needs to be divided into one ... to up to 7 folders as shown in the case of the violin.

STEP 02: Open up ras' folder where the database is being kept. This database is online and alive hence making it possible to link from the internet. Looking at the contents of T:\COIT\MYDEPT\cs\ras\MUMS as one can see there is no marimba folder. Automation of this Step is moot as it takes only a second for a human to open up a folder.

STEP 03: Create a set of marimba folders in ras' database. Automation of this Step is moot as it takes the same amount of time to tell a creating/renaming program - what the new folder should be named. Here I decided to not automate this process.

STEP 04: Create a set of marimba folders in the MUM's temporary folder for the purpose future automation. Again, as in step 04 - automation of this Step is moot as it takes the same amount of time to tell a renaming program - what the new folder should be named. Here I decided to not automate this process of simply creating and naming the folders:

STEP 05: Open Sound Forge 8.0 and open up the two newly duplicated and created marimba folders. Clearly one can see the distinctive wave forms of the single stroke versus the crescendo rolls. I simply delete all the rolls in the single stroke folder and delete all single strokes in the crescendo folder. See Step 06. Automation of this step can be done. Once could make a little C program to distinguish between the abruptness of the single stroke compared to the Gaussian-like attack of the crescendos, but all this time making such a program would be wasted when we went to another instrument, such as 'bowed" versus merle on a violin.

STEP 06: Check to see that file separations are properly executed and then save both files. Automation of this step -- moot - I simply listen as a human being to see that the sound have retained their integrity.

STEP 07 AUTOMATED: We use the following windows script to automate 14 procedures illustrated in STEPS 07 through STEPS 09:

#z::
Send, {CTRLDOWN}x{CTRLUP}{ALTDOWN}f{ALTUP}n{ENTER}
sleep 100
Send, {CTRLDOWN}v{CTRLUP}
sleep 100
Sleep, 100
Send, {ALTDOWN}f{ALTUP}a
WinWait, Save As,
IfWinNotActive, Save As, , WinActivate, Save As,
WinWaitActive, Save As,
Send, {TAB}{TAB}{TAB}{TAB}{TAB}{TAB}{TAB}{TAB}{TAB}{TAB}{DOWN}{UP}

It should be noted at this point, even though these steps are automated, there is, as will be demonstrated, certain stages of the aforementioned automation that require human intervention, checking and lastly decision making. As seen here in STEP 07, the first stage of automation. The start point of the wave is not the first instance where there is a variance above a certain threshold form no sound to sound. Rather, it requires a human ear to distinguish "noise" or dirty sound of the person playing the instrument as he/she shuffles there feet, lets the instrument touch their clothing or allow the bow to accidentally touch the violin. Of course a C program could be easily made to distinguish - no sound to sound, but a human's ear is need to distinguish noise to viable, database sound. Automation of this step is already implemented after a human distinguishes noise from sound

STEP 08 AUTOMATED: Creating new window: Already automated

STEP 09 AUTOMATED: Naming the selected and extracted sound in the new window: This step takes the longest time however, it is the most important step, naming the tag for the sound: A - is the note, 2 - is the octave. Right here, one could say - can't we scan in the MUM's catalogue, run it through an array and have it parsed onto the labeling? Yes - this is indeed a good idea, however, the MUM's database, and the Oginara database have incorrectly labeled many of the pitches. These are due to typos or simply not being musicians. IN short, I listen to the sound and compare it to the pitch on my keyboard. I have found seven mistakes so far. A computer would not have found these and would wreak havoc in our system if not spotted.

Here I check the A2. Once good, I hit CNTL-V and pastes the "marimba_crescendo_M"

STEP 10 AUTOMATED: Looking at the window T:\COIT\MYDEPT\cs\ras\MUMS one sees we have one file of the A2_marimba_crescendo_M.au successfully inserted into Dr. Ras' database.

STEP 11: The above steps were all done to only create one of the seven files in marimba: crescendo roll, one needs to implement the batch file to continue extracting the other six files in marimba: crescendo roll. This is a batch file, automated as much as humanly possible.

STEP 12: Continuing, another observance of human intervention, sometimes MUMS will name the C but with the incorrect octave. In this case it was 3 but sometimes MUMS incorrectly keeps the previous octave. Only by physically playing the note on the keyboard, listening to it and then checking it can our Database have the integrity we demand.

STEP 13 : All seven files have now been converted.

STEP 14: Checking them in Sound Forge, we make sure the beginnings and endings are good. Note this is only the seven for crescendo. the single strokes still need to be ripped.

STEP 15: As one can see - Sound Forge includes the SFK (Sound Forge Peak Data Files) and SFL (the cue points) in a separate file format but as files in the database. These will not be used by the database. Note there is no SFK or SFL on the single stroke - because they have not been ripped at this point.

STEP 16: Note these are only seven of 56 sounds for Marimba Times two (crescendo - single note) = 112 sounds for Marimba, so fare we have shown how to do seven of the 112.

STEP 17: The (8 x 2 = 16 x 7) remaining rips to be performed for Marimba