Retro sounding speech synthesizer

[antibyte]

Just found this one :

http://www.societyofrobots.com/member_t ... s/node/211

Lets you add a text to speech engine to your games !
Eats 12 K flash and sounds like the old 80s speech synthesizers

Have fun
antibyte

[uze6666]

Thats *very* cool thanks for the link. I've used a Magnevation SpeakJet recently (in a robot, the Uzebot ) and its footprint looked just like a mega88 or some PIC chip. Now I'm pretty sure it was running such a software. Sounds good though nothing earth shattering. Anybody want to port it to Uzebox libraries?

Uze

[havok1919]

Thats *very* cool thanks for the link. I've used a Magnevation SpeakJet recently (in a robot, the Uzebot ) and its footprint looked just like a mega88 or some PIC chip. Now I'm pretty sure it was running such a software. Sounds good though nothing earth shattering. Anybody want to port it to Uzebox libraries?

I think the early SpeakJets were PICs, IIRC. They certainly could have changed over the years!

-Clay

[antibyte]

The Speakjet uses an other aproach.
It synthesizes speech using a vocal tract model replicated by noise and pulse generators
and filters.
Speech is generated by constantly feeding the generators and filters with parameters.
This approach needs alot of calculations, but only very little memory.
The Synth posted above uses prerecorded samples, requiring more memory but only very little of CPU time.
This one should perfectly fit Uzebox.
The only problem is, the documentation states that its okay to change the Source but not to redistribute
it without permission from the Author.
But maybe the Author gives permission to add it to Uzebox, if someone asks him.

Greetz
antibyte

[havok1919]

Might be fun to disassemble something like the old "Software Automated Mouth" (S.A.M.) that was on the Atari 400/800 and other 8-bitters and see how that worked. Probably phoneme based. It was definitely "swedish robot" sounding like the phoneme chips-- SP0256, etc. The earlier Orator chip (SP0250) sounded 'better' but required more CPU overhead... I might have to read the MAME code a bit and see how it actually works. I remember Frank Palazzolo poking at it a long time ago and getting things pretty well reverse engineered.

The Synth posted above uses prerecorded samples, requiring more memory but only very little of CPU time.
This one should perfectly fit Uzebox.

I guess since it's voice-grade anyway we could just spit out a simple ever other scanline time or something for a ~7.5KHz sampling rate. Could get away with ~131 bytes or so of sound buffer that way and just generate the next audio frame in vblank...

-Clay

[antibyte]

I asked the Author, Clive Webster, about the applied restrictions of his Speech Synthesizer code.
Here is his response for clarification :

Hi,

Thanks for the comments.

Yeah the reason for the restriction is that I am currently in conversations with someone regarding them licensing the code. It may, or may not, come to anything. But if it does - then they will probably want me to remove the source code from the web. Until this is resolved - I'm not posting any updates to the code to the SoR forum. If it comes to nothing then I may give permission to Uzebox and put it into the public domain.

Have looked at your Forum post - and it may be useful to know that the current phoneme sample rate is around 5.5kHz.