A statistical examination of the predictive audio buffering algorhythm in case of musical content

Gabor Gerenyi
Overmind Ltd.

 Performance parameters:

 - Number of simultaneous channels

 - Number of punch-in points

 - Compressed/uncompressed

 - Bit depth

 - Sampling frequency

 Theoretical maximum bandwidth of a full stereo project

 - 96 channels

 - 32 bit depth

 - 192 kHz sampling frequency

 - Uncompressed

96 x 4 x 2 x 192.000 = 147.456.000 byte/sec

- Punch-in points – can be anything between 1 and a few thousand

 - Great help in saving bandwidth

 - Great help in predicting and preparing content

Finding and prebuffering punch-in points in the future is a key element:

- saving resources

- making the system more reliable

- making the system more usable

 Consumer streaming services:

 - Soundcloud

 - YouTube

 - Vimeo

 - TikTok

Linear time model…

- starts from the very first punch-in point

- buffering a fixed sized frame consecutively

Pro-grade application streaming

 - Many streams at the same time (channels)

 - Non-linear time model

 - Reading ahead in time with jumps to future punch-in points

Multiple buffers are preloaded with the next available punch-in point data...

Musical audio content

 - High number of used channels

 - Relatively small number of channels at the same time

  •  Many versions of the same musical line

  •  Same instrument needs different effects

  • Anyway, high channel number would be very confusing musically

Prebuffering strategies

 - Absolute

 - Time-frame limited

 - No-buffering – not a real option

Buffering cycles are easy to track on the next slide

Thank you!

Gabor Gerenyi

Overmidi Ltd