In general, a sample workbook (wbk file) is placed in ROM and only the necessary parts are DMA transferred to RAM as needed for a given frame. For this reason, a buffer area must be secured in RAM ahead of time. The area secured in RAM is specified at initialization time by syn_dma_buf_size (the size of each buffer) and syn_num_dma_bufs (the number of buffers).
The optimal values for these two parameters will differ depending on the contents of the songs and sound effects so there is no way to clearly define them, but you can use the following general measures to make approximations:

Buffer size (syn_dma_buf_size)
First of all, we will begin by considering the buffer size needed to DMA transfer one channel's worth of data. Since the sample data is treated as 16bit monaural sampling, each sample point has a size of 2bytes (i.e., 1 (monaural) x 2 (16bit =2byte) ). Therefore, to playback audio with a 32000Hz sampling rate at the same pitch as the original audio, you need 64000 bytes per second, or 1066.66.... bytes per frame. However, this sample data is compressed to around 1/4 its size by ADPCM conversion, so you actually end up with 267 bytes (rounded up).

In the above example, the conditions were set to playback audio at the same pitch as the original. This is because you would use more bytes of data to playback at a higher pitch, and less data to playback at a lower pitch. Sound Tools can playback audio at pitches as high as 1 octave above the original audio, so the maximum amount of data needed would be double the size.

Number of buffers (syn_num_dma_bufs)
Basically, you need one buffer for every channel. However, not all data is DMA transferred every time; rather, data is cached temporarily in buffers in order to cut down on the total number of DMA transfers. As a result, you need more buffers than there are channels. With the MUS library, in order to store 1 frame's worth of data you need at the very least double the maximum number of channels (in this minimal configuration there are no channel cache hits).
Based on the above considerations, the following formulae can be used to approximate buffer size and buffer number:

(Buffer size)  =   (maximum sampling rate)    ÷   (number of frames/second NTSC(MPAL)=60, PAL=50)    ×   (maximum ratio to original pitch (approx. 2))    ×1   (monaural (fixed))    ×2   (16bit sampling (fixed))    ÷4   (average compression rate due to ADPCM conversion (fixed))

(No. of buffers)  =   (maximum number of channels)  ×   (maximum value of(2 - cache hit ratio))

Although , and can be quickly determined for these calculations, deciding which values to set for and is more problematic. We recommend setting a maximum value of 2 for unless you are being careful during the creation of sounds and sound effects to make sure the audio does not playback at a higher pitch than the original. Once is set you can determine the size of the buffer. As far as the number of buffers is concerned, you need to search for the optimal value by initially setting =2 and then gradually lowering the value to find the minimum value for which noise is not generated. Note that you will need to have some extra buffers, since the necessary number of buffers can change during playback as songs and sound effects are combined.

Making the buffer size larger than the calculated value has the effect of decreasing the number of DMA transfers. However, if the buffer size is too large, the time needed for DMA becomes longer, and concerns about being on time for the start of waveform synthesis arise. Consider a size of around 2 kilobytes to be the limit. If you set the number of buffers greater than the needed number (e > 2), the only thing that will happen is that you will waste memory.

Note that the values obtained here are strictly a yardstick, and things are actually a bit more complicated because of the characteristics of ADPCM. More specifically, the buffer size can in fact be reduced by around half, but if you shrink the size to the lower limit, you are going to witness a huge increase in the number of DMA transfers. We thus recommend using the standard when deciding on your values.