Transylvania, a new version using TMS5220 synthetizer

[+retroclouds]

Jespol,

 

Could you post the one of the source files (the output from QBOX with the BYTE data)? I'll give it a try. I'm sure it should work just fine.

 

I looked at the .bin files that you posted, but I can't/don't know how to convert them into a file that I can work with. I really need source code.

 

Mark

 

This is a little utility I wrote to convert a binary file to source code data statements.

I remember using it to do some tests with QBOX

 

 

bin2byte.zip

Edited September 30, 2012 by retroclouds

[jespol]

Ok.

I have attached a new file with the ASM code which I used to test all the LPC samples.

sfx_asm.zip

[rocky007]

yes it's working !

i think something was bugged in my bin to byte conversion

 

voices are good, but samples like wolf are little bit strange...

 

 

Edited September 30, 2012 by rocky007

[sometimes99er]

Sounds like emulation - only !?

[rocky007]

you're right, i have to try on real TI...but actually i still didn't connected sound on my TI99 ( F18A)..

here it is the archive...

SYNTHTEST.zip

Edited October 1, 2012 by rocky007

[jespol]

You're right... some samples are not present in the game because the real sound is worse, but the emulation sound was good !!!

In fact emulation of the 5220 is not totally reliable or it does not emulate some bugs of the TMS52200 (I vote for the second solution).

But all the samples in the game sound well (or not too bad).

 

You have the complete collection of samples we have generated from Qbox, but only some of them have been retained (cause of bad sound, french voice, or simply unused).

I dont't think that other elements could interrest you because the video processor TMS3556 is unknown in your country (and in the rest of the world), and the TMS7000 serie CPU has never been used in a TI computer... but it is a very interesting microcontroler (not fast but easy to code and it integrates all we need in a single chip).

 

I heard sounds you obtain with a Ti99. It is clear that the revision of the TMS5220 chip is different or the coefficient table in ROM is different. The result on an Exelvision machine is far better than the emulator, sounds are clear (with some exception) with good bass (Exelvision has added some kind of filters).

Edited October 1, 2012 by jespol

[rocky007]

so, if i understood well :

 

TI99 don't have TMS5220 but the previous version, TMS5200, labelled CD2501E-NL

and it's a huge difference in sound quality !

 

Unfortunally, i didn't found the CD2501e-NL datasheet to see if it's possible to remplace the original CD2501E-NL by the latest TMS5220 & roms, but i found a partial answer on WIKI :

 

TMS5220 (AKA CD2805E?): Improved version of the TMS5200, pin but not function compatible (has new LPC tables and a new chirp table); used on (later) Apple II Echo 2 cards, (rumor) on the very last run of TI 99/4A speech modules.

Edited October 2, 2012 by rocky007

[jespol]

And we used a modified coefficient and chirp tables with QBOX, not the original one !

Last revision of 5220 is embedded in all exelvision computer (Exl100 and Exeltel): Exeltel allows to change the frequency (male/female voice) and adds a 6100 ROM chip (with many french words and some sounds).

[Willsy]

WOw! it sounds like a really great little computer! I want one!

[jespol]

Yes... but with one major problem: TMS7000 CPU uses SRAM and not DRAM, and SRAM was very expensive... so the computer was built with only 2ko of SRAM and 32 or 64ko of DRAM for Video purpose. And the communication with the DRAM is via the TMS3556 video controller... and it is slow, very slow ! Hopefully some SRAM extensions exist (16ko cartridge and 32ko embedded in the floppy device).

 

The other main problem is elsewhere. At the same time appears a brand new computer: Amstrad CPC with its known great success in european countries... and in 1986 (when exeltel appears), the 16/32 Atari computers push the 8bit platform out.

[+retroclouds]

Tried the thunder3.tasm file. Quite impressive that you can use the Speech synthesizer for other things than speech as well.

Attached is an 8k cartridge binary you can use to give it a try. Ofcourse, you need a speech synthesizer

thunderc.bin

[Kurt_Woloch]

So, how is the new speech conversion program coming along? Would it be possible to alternatively have it work with a different coding table? As far as I know, the architecture of the TI-99's speech synthesizer is pretty much the same as the TMS5220, it just uses a different coding table with is partly supported by QBOX, and slightly different parameters. The parameters would be as follows:

 

- The chip type would be a TMS5200/TMC0285 (look for SUBTYPE_TMS5200 in the file tms5110r.c which is part of the source code of MESS, there you will find the correct coding table).

- The sampling rate is 8000 Hz, and the frame length is 200 samples, so 40 frames occur per second. This is actually unsupported by QBOX which assumes 200 samples per frame for 10 kHz, but 160 samples per frame for 8 kHz, which in both cases would result in 50 frames per second. In order to process speech correctly with QBOX, in addition to supply the correct coding table (I have one if someone needs one), you'd have to first convert the source material to 8 kHz, then speed it up to 10 kHz and save it as that, feeding the result into QBOX and processing it as 10 kHz. It would be fine if your program would allow converting to the TMS5200 without using such tricks.

[+retroclouds]

oh, that isn't me. Is that mizapf working on something new?

[+mizapf]

Is that mizapf working on something new?

 

What mizapf? Me mizapf?

 

The emulation in MESS uses the TMS5200, as found in the real speech synthesizer. The TMS5220 sounds pretty different.

 

I'm not involved in any speech conversion tools.

[Kurt_Woloch]

No, it wasn't Mizapf. I answered to the text quoted here:

 

We have found sources of a modern encoder for LPC string, and we will try to adapt the output (and the algorithm) for TMS5220... I hope this new converter will enhance human and non human samples.

 

If anything, Mizapf is involved in speech output of M.E.S.S., as he's (or was?) the maintainer of the TI-99 driver of M.E.S.S. To that end, I've just downloaded M.E.S.S: 0.155 (the current version) to check its speech support, and it doesn't speak at all... maybe I just need a new set of the TI-99 ROM's, or I need to do something else. But it doesn't list a speech synthesizer under the "sound" devices anymore, so I suspect it doesn't support speech anymore.

[+mizapf]

I'm indeed the current maintainer of the TI-99 drivers in MESS (note that we dropped the dots a while ago). Speech is certainly available in the current MESS release; you may have outdated ROMs or you did not activate speech. Go to ftp.whtech.com, folder /emulators/mess/roms, download ti99_complete.zip and unzip it into your rompath. Do not unzip the zip files inside that archive.

 

In order to enable speech you must plug in the speech synthesizer into the P-Box. From the command line, add

 

 

-peb:slot5 speech

 

if you want the speech synthesizer (plus adapter board) to be put into slot 5. Note that the slots have different options (slot 8 is reserved for the floppy controllers).

 

Michael

[Kurt_Woloch]

OK, thank you! I managed to get it to work, and it sounds pretty nice now. I only think there's a bit of sampling noise on the speech... maybe a slight low-pass filter would do it good. It might already be enough to smooth out the transition of one sample to the next (I mean the samples that come out of the speech synthesizer at the 8 kHz rate).

 

I'm indeed the current maintainer of the TI-99 drivers in MESS (note that we dropped the dots a while ago). Speech is certainly available in the current MESS release; you may have outdated ROMs or you did not activate speech. Go to ftp.whtech.com, folder /emulators/mess/roms, download ti99_complete.zip and unzip it into your rompath. Do not unzip the zip files inside that archive.

 

In order to enable speech you must plug in the speech synthesizer into the P-Box. From the command line, add

-peb:slot5 speech

if you want the speech synthesizer (plus adapter board) to be put into slot 5. Note that the slots have different options (slot 8 is reserved for the floppy controllers).

 

Michael

[tr1k1]

So, did the "new" encoder ever happen?

I'm also looking for a way to directly convert SP0250 to tms5200/5220 without simulating the SP0250 and re-encoding for the tms.

[deladriere]

So, did the "new" encoder ever happen?

I'm also looking for a way to directly convert SP0250 to tms5200/5220 without simulating the SP0250 and re-encoding for the tms.

 

Check this out ! : http://atariage.com/forums/topic/260857-introducing-bluewizard-qboxpro-replacement-speech-analysis-tms5220-tool

[Mehridian Sanders]

Did this port make it to the TI-99/4a? I must confess, my father (having the power of controlling what was bought) purchased a C64 and Transylvania. I do remember playing it and I would like to do so again on my one true love... The TI-99/4a.

[senior_falcon]

With the exception of the speech, Transylvania is a perfect example of what I hoped would be written for The Missing Link Graphics Adventure, which lets you load pictures to the upper 2/3 of the screen and have text in the lower third.

[Kurt_Woloch]

OK, I did a bit of an experiment here in order to see how well the graphics would translate to the TI-99. I actually meant to convert the first picture that's seen in the video, but since I wanted the highest possible quality in the source material, I started with the Amiga version. But despite the Amiga supporting 32 (or more) colors at once on screen, it still dithers the graphics. And then I noticed that it actually doesn't start with the picture shown in the video, but in the woods by a stump, so that's the picture I converted. To clean it up, I first un-dithered it by using colors that are the average of the two colors that got mixed and filling the respective area with that color. Actually, for the stump picture, still only 8 colors are needed (plus the colors for the non-graphics area). I then ran the resulting picture through my converter program to see what the colors converted to, and I saw that three of the areas consisted of mainly one color with a few differently colored pixels sprinkled in, so next I filled those areas with that color (which is one of the colors supported by the TMS9918A) so that the areas remain solid.  There are, however, two dark areas which remain dithered because the TMS9918A doesn't support such dark colors. What's also evident is that there are problems with the black lines if they are separating two differently colored areas. But all in all, I think it doesn't look too bad, so I've attached the resulting picture which respects the resolution and colors of the TMS9918A as well as the assembler code for the pattern generator and color tables you need to display the picture on it.

In reality, from the way the pictures are drawn, I guess the author of the original game didn't really store bitmapped graphics, but rather "draw" and "fill" commands, where the fill also supports dithered colors, and then there are some shapes stamped over the drawn graphics. Thie probably saves some memory vs. bit mapped graphics.

Transylvania_TMS9918_asm.zip

[Kurt_Woloch]

Just for comparison, here's another, more complex picture I converted from the Amiga version without first touching it up.

Transylvania_2_TMS9918.zip

[Kurt_Woloch]

And this is a conversion of the title picture of the Exelvision version, taken from the author's homepage because I suppose that picture isn't part of any other version.

Transsylvania_Title_TMS9918A.zip

[GDMike]

Well.. some day we might get a video enhancement...grin