[ Home | Plans | Meetings | Members | Search | Papers | Links | CyberWear | LocoSoft]
Storage Space Breakdown for the LocoMedia System - in progress!
Its a little too early to finalise the storage space budget, but here are some facts
and assumptions that will effect the final numbers:
- We have 80MB of flash storage available.
- Standard Linux requires 60MB, but Henk guarantees to strip this down further to 40MB.
More might be possible thereafter.
- Assume avg note is 20wds at 0.6sec/wd => 300 messages per hour.
- Assume that 300 messages is sufficient for trials and demonstrations of LocoMedia, and
that this includes all system reponse messages.
- Assume Domain Manager requires ?? for executables and domain data. Ed - TBD
- Assume FootBridge protocol stack requires ??? James - TBD.
- Mike's components...
- Alistair's audio components, and some space set aside for a cheap and cheerful TTS.
- Reserve 5MB for future applications and as safety margin.
How much space do we need for the HP speech engine's training data?
- Assume that we store HM model for only one user at a time on the CyberJacket.
- Set aside space for user model sufficient for the currently defined grammar, plus half
again for future extensions => 20 words. Alistair?
- Set aside space for user model sufficient for 20 domain names, averaging 5 words =>
- Set aside space for user model sufficient for 20 place names, averaging 5 words =>
- Assume roughly 200 words total at roughly 1k per word (Thanks, Akos) => 200kbytes
total for vocabulary.
- Assume <200kbytes for speech engine itself (non-debug version of continuous speech).
To be confirmed on Linux system. Alisair?
So total storage space required for speech engine and HMM data should be <400KB.
The Speech coder effects how much space we need to store 300 messages / 1 hour of
audio. For the mean time lets assume that we will be using a speech coder and see how they
do with ambient sounds (pigeons cooing, brass band playing, traffic tooting, crowd noise,
hot air balloon burners). We might consider using Roger Tucker's (HP) parasitic converter
if need be to preserve high frequency quality.
Two speech coders are available:
- GSM: 6MB/hour. Faster encode and better quality than CELP?
- CELP: 2MB/hour
- What excecutable size?
- What RAM requirement?
- What input and output latencies on our platform?
- What audio quality for speech and for ambient sounds?
The material displayed is
provided 'as is' and is subject to use restrictions.
For problems or questions regarding this web
contact Cliff Randell.
Last updated: January 14, 2000.
ęCopyright Hewlett-Packard 1997-2000.