nRF51822 as co-processor for Microbit

[Sean]

I am working with the BBC on the Microbit project. bluetooth nRF51822 has an ARM Cortex-M4F CPU which is MUCH more powerful than the ARM Cortex M0 being used as CPU. Is it possible to use this chip AS the CPU? The main use I foresee is to decode CELP for Audiobooks. Since encoding doesn't need to be real time, an exhaustive search can be used for the encoding step, enabling CELP to match ACELP in sound quality. ACELP is used by vendors like Audible and if the ESCELP (exhaustive search code excited linear prediction) can produce equal quality, it would allow audiobooks, Q&A to follow written work and maybe even testing using a formant-testing algorithm to distinguish the numbers zero to 9 (multiple choice). Lessons could even be recorded & saved to be listened again by the pupil.

I see the Microbit as an opportunity for a total sea change in UK teaching which would mean 1 million new Microbits per year so Nordic would be able to secure a much larger order.

With thanks, Sean Dunlevy

[Sean]

I am - I was pointed to the page and read from the top (duh). So, I have 16K of RAM & 128K of Program ROM (well, flash-RAM, but you know what I mean).

Has there ever been an asynchronous compression scheme in which the encode doesn't have to be real time. LD-CELP (16Kb/s) gives reasonable quality and because the frames are so short, going between voiced & unvoiced will be less jarring.

Frankly, without a proper spec on how the Bluetooth & main CPU link up (among other things), I don't plan to write a custom speech codec.

Oh - I see that Nordic themselves favour ADPCM. Very low complexity BUT can be improved. Note that there is an unused code (0 & -0 are the same, as a multiplier).

Appreciate the time, friends.

[JohnBrown]

Depends which version of the nRF51822 is used on the Microbit. The later versions have 256K Flash and 32K RAM, as far as I know, and even the Rev 2 has 256K Flash, but only 16K RAM. Maybe someone from Nordic could say exactly which chip is used?