Rockbox mail archive
Subject: Re: an idea/suggestion
From: BlueChip (cs_bluechip_at_webtribe.net)
> > If synthetic speech is using up the 2 Mb ram, it may need more memory for
> > processing the MP3s.
> > I do not understand which data is loaded where yet, so I know that this
> > might be wrong.
>You are missing the real problem: We can only play mp3 data. Even if we
>could synthesize speech, we need to mp3-encode it before it can be heard.
>The 12MHz cpu is much too slow to mp3-encode in real time.
presample all your phonemes (a debatable 47 for English if my memory serves
me well) compile to mp3 on a PC, pre-load all phonemes to memory (maybe
this is the use for the extra few MB of RAM) then fire them off with a
text-to-phoneme algorithm - we'll be a few years back on quality, but the
principle is fairly simple.
Of course, once we get the SDK for the Micronas MAS 3587F (and MAS 3507D
for "player" support) we will have WAV output :) Yes, I know we will not
be playing 44,100Hz 16bit samples, but for speech this is a superfluous
issue anyway. Once we are playing WAV files, we will also be able to use
far more sophisticated text-to-speech algorithms. Cool huh?
Page was last modified "Jan 10 2012" The Rockbox Crew