April 26, 2015

zxsp 0.8.pre26: hE(ll)(oo) k(ee)(OO)

Got it?

"Hello Kio !"

Yeah, my Spectrum speaks.

Inspector for the Currah µSpeech digital speech unit
Thanks to the work of Joe Zbiciak and Frank Palazzolo, who had reverse engineered the SP0256 speech processor, i was able to implement an emulation of this chip and then of the Currah µSpeech interface in zxsp.
The SP0256-AL2 contains a cascade of six 2-pole filters, a rom with filter coefficients for British/American language and a micro sequencer, which feeds these values into the filter coefficients. All digital. The output is an 8 bit PWM.

Despite of the very comprehensive informations about the SP0256 chip on their website there were still some bits not clear to me. I filled in the gaps by own tests and looking at what other people did. (not too many of them…) And i emulated the double RC filter after the PWM output. Yeah!

I added an option to switch between the original 8 bit sound and the unrestricted approx. 14 bit output, but i can't hear any difference. :-)

At the bottom of the inspector widget i added a scroller which displays the spoken allophones.

Though most allophones are spoken as expected, i believe some are emulated wrong, e.g. "1" and "Y" sound a little bit like a "bass bump" and "?", spoken as "kw(ee)r(ee)", sounds more like "kneenee". But there is little i can do about this.

I found that there are three pairs of allophone tokens in the µSpeech unit which refer to the same allophone: "(ck)" and "(gg)" which refer to KK2, "(oo)" and "(eau)" for OW and "(aa)" and "(ay)" for allophone EY. On the other side, tokens for two allophones are missing: DH2 and GG2. Given the fact, that "CLEAR" and "CLS" are both spoken as "CLEAR", the µSpeech seemed to be released a little bit fast these days.

Technically the Currah µSpeech isn't a Z80 peripheral. It's more like a 6502 add-on. It uses memory-mapped i/o! I had to extend my Z80 emulation to handle that sort of things…

Final problem was the rom switching for this little beast: It toggle's between it's 2k rom and the ZX Spectrum's internal rom when a RST7, that is, when the regular timer interrupt is executed. Actually, the way i handled rom switching up to now could not handle this: It detected an opcode fetch at address 0x0038 in one rom, switched the rom and restarted the CPU at this address. Then it detected an opcode fetch at address 0x0038 in the other rom, switched the rom and restarted the CPU at this address. And so forth.


A final word about the Blog Editor: It's getting worse every day i write my next post. It's really so bad now, that i write my blog post in a text editor on my Mac and copy&paste it into this editor.

No comments:

Post a Comment