+Vorticon Posted July 27, 2015 Author Share Posted July 27, 2015 Yup, I was correct. With only a minor modification to the prototype board, I was able to speed up the image transfer rate over 35x, going form 62 minutes to 1min 42secs! And I have not yet implemented the byte packing scheme Tursi suggested, which should give me an additional 250% speed increase at a minimum (accounting for the extra processing overhead for packing and unpacking). This should bring down the transfer time to somewhere around 41 seconds.Here's what it looks like currently: https://youtu.be/2STv2FUbytI 5 Quote Link to comment Share on other sites More sharing options...
+Ksarul Posted July 27, 2015 Share Posted July 27, 2015 Nice progress, Vorticon! Quote Link to comment Share on other sites More sharing options...
Opry99er Posted July 28, 2015 Share Posted July 28, 2015 Wow... That is some breakthrough!!! Well done! Quote Link to comment Share on other sites More sharing options...
+Vorticon Posted July 28, 2015 Author Share Posted July 28, 2015 So I implemented the color depth reduction scheme as suggested by Tursi, and it brought down the image transfer time from 1'45" to 1'13", a 70% boost in speed. This was kind of disappointing as I was expecting something in the order of 250%. Bitwise manipulation in Python is apparently not terribly speedy, and there is the added overhead of unpacking the bytes on the TI side. The last available optimization is creating a lookup table for the color conversion on the TI as this is currently taking up a good portion of the processing time during the transfers. More to come Quote Link to comment Share on other sites More sharing options...
+Vorticon Posted July 29, 2015 Author Share Posted July 29, 2015 OK, with the color look up table in place, the transfer time is now down to 47 seconds. I think this is the best I can probably get out of this setup. There are probably some minor code optimizations I could do, but I doubt it would shave more than a couple of seconds at best. I am not terribly happy with the color rendering however with the look up table as compared to the original color distance calculation method, so I will need to tweak the table a bit to get better results. I'll post another video of the final result once that is done. Overall though, a success It only took a year of intermittent experimentation... 1 Quote Link to comment Share on other sites More sharing options...
Opry99er Posted July 29, 2015 Share Posted July 29, 2015 Very cool, Vorticon... Very cool indeed. 'Vorticon, his eyes open' Quote Link to comment Share on other sites More sharing options...
Tursi Posted July 29, 2015 Share Posted July 29, 2015 It's possible that the color depth at 332 is not good enough for the matching you want to do - it will definitely look different. I used it a bit in the DOS days and it's really best qualified as "ok". You /could/ skew the color matching table a bit, rather than going for pure mathematical matching, since you know the source image has color shifted. There's also a trade off of how much you want to 'cheat' by using the PI's processor (ie: if it did the color matching, you'd have the full original source image). 47 seconds sounds about what I'd expect though, so that was a success! Congratulations! Quote Link to comment Share on other sites More sharing options...
Omega-TI Posted July 29, 2015 Share Posted July 29, 2015 Very cool, Vorticon... Very cool indeed. 'Vorticon, his eyes open' Tamarian? Quote Link to comment Share on other sites More sharing options...
Opry99er Posted July 29, 2015 Share Posted July 29, 2015 .....when the walls fell.... Quote Link to comment Share on other sites More sharing options...
+Vorticon Posted July 29, 2015 Author Share Posted July 29, 2015 It's possible that the color depth at 332 is not good enough for the matching you want to do - it will definitely look different. I used it a bit in the DOS days and it's really best qualified as "ok". You /could/ skew the color matching table a bit, rather than going for pure mathematical matching, since you know the source image has color shifted. There's also a trade off of how much you want to 'cheat' by using the PI's processor (ie: if it did the color matching, you'd have the full original source image). 47 seconds sounds about what I'd expect though, so that was a success! Congratulations! Thanks Several times I came very close to pulling the plug on the project, but you guys have been a great source of help and have always come up with suggestions to get me through each thorny problem. As far as the image processing aspect, what I'm currently doing is first turning off any pixels that are close to black in any 8-pixel block, then averaging the combined RGB values for each 8-pixel block, and comparing that average to the RGB average for each TI color. It does not get much cruder than that, but it's fast. The problem here is that certain colors are very close to each other and in one case has exactly the same average, so I plan on evening up the distribution a bit and see what I can get. It's going to be purely trial and error here though. And yeah, I really want to showcase the TI's abilities here. If I wanted to cheat, then I could have the Rpi run a full dithering algorithm on the entire image and simply just pump the converted TI color values to the TI to get great looking images a la Convert9918, but where's the challenge in that? Once I clean up the color rendering a bit, my next challenge is going to be to get the TI to do a simple object recognition, likely just a colored square on a flat surface (at least to start), no matter where it is on the image. I still have to do some of research on this. I found an old article in a Robotic Age compilation that looks very promising. 2 Quote Link to comment Share on other sites More sharing options...
+Vorticon Posted July 30, 2015 Author Share Posted July 30, 2015 I've been trying to tweak the color table, and I could not achieve a substantial improvement in the image, so I decided to try converting the RGB image to black and white for comparison. Below are the results: I think B&W is a definitely clearer than the color version, and I may stick with it. Sabre Wulf cart Yours truly 3 Quote Link to comment Share on other sites More sharing options...
+Vorticon Posted September 6, 2015 Author Share Posted September 6, 2015 You don't have an MBX system? No worries! Here's a fun replacement using late 80's technology with the Archer VCP200 voice recognition IC from Radio Shack which will allow you to control virtually any game using voice commands in lieu of the joystick. It is a speaker independent system that does not require training. However, it is only limited to a few words, although still quite usable. https://youtu.be/uBc2XhaJ0-g 3 Quote Link to comment Share on other sites More sharing options...
Omega-TI Posted September 6, 2015 Share Posted September 6, 2015 Quote Link to comment Share on other sites More sharing options...
Dexter Posted September 6, 2015 Share Posted September 6, 2015 Nice project, I like it! Quote Link to comment Share on other sites More sharing options...
+Vorticon Posted September 6, 2015 Author Share Posted September 6, 2015 Thanks I'm going to create a PCB for it and print an enclosure with my 3D printer. I'll also add a fire button. I need to find a good headset microphone like the MBX one. The final product will be a universal voice activated joystick. It will make a nice addition to my eclectic experimental hardware Quote Link to comment Share on other sites More sharing options...
+Vorticon Posted October 28, 2015 Author Share Posted October 28, 2015 Here's the home made PCB for the voice recognition project,all drilled out and ready for the parts. I should have a fully assembled prototype by Faire time this weekend And now a rant: I miss my Radio Shack store! Say what you want about the management of RS over the last decade, it was still the fastest way to get your hands on common parts when in a pinch and unable/unwilling to wait a week for an online order to arrive. With the closing of a large number of stores, now the closest RS store to me is 25 minutes away without traffic whereas it used to be only 5 minutes. Today I needed a 7805 regulator and a couple of switches for the project, so I drove over there and got what I needed without fuss. The store clerk had absolutely no idea what I was doing, but I was used to that by now anyway. You really can't beat that kind of accessibility, not to mention that I love to rummage through the store every time I visit looking for interesting finds. I keep hoping that RS will rebound with the resurgence of the DIY movement... 4 Quote Link to comment Share on other sites More sharing options...
+Vorticon Posted October 29, 2015 Author Share Posted October 29, 2015 Here's the assembled and tested board. There was a missing trace in the design, easily fixed. I also trimmed the edges a bit to make it fit into the project box. Works great! Tomorrow I'll be packaging it inside a project box 2 Quote Link to comment Share on other sites More sharing options...
Opry99er Posted October 29, 2015 Share Posted October 29, 2015 That is flippin fascinating. Quote Link to comment Share on other sites More sharing options...
Omega-TI Posted October 29, 2015 Share Posted October 29, 2015 (edited) Vorticon, I really like this last speech recognition project of yours, but while investigating the possibility of re-creating it for myself, I've hit a dead-end when it comes to sourcing the Motorola chip. It appears to be 'unobtainium'. In fact I've not been able to even locate a datasheet on this IC. I have found that it's a "bit-masked programmed" 6804 microprocessor. So, if a blank M6804 can be located, is there any chance of reading your chip and reprogramming a blan device, or is it part of the manufacturing process, which means we're all screwed and can just be envious of your new TI toy? ** EDIT ** Could you please scan the documentation that came with it and post it to the forum? Motorola 6804 - MCU Manual -Sept 1985.pdf Edited October 29, 2015 by --- Ω --- Quote Link to comment Share on other sites More sharing options...
+Vorticon Posted October 29, 2015 Author Share Posted October 29, 2015 Yes I'm afraid it's pretty scarce, but I was able to secure a spare one on Ebay a couple of months ago. I think they pop up every once in a while. I'll keep an eye on one for you . There is no published datasheet for the VCP 200 as it is considered proprietary and the voice recognition code and data are permanently burned in ROM. What I do know is that it is an early Motorola microcontroller with 1.5K of ROM and a few bytes of RAM. As to cloning the chip, I'm not sure if that is feasible but I am definitely not an expert here. ksarul may be the person to answer this one. I'll try to scan the documentation tonight if I have the time otherwise I'll do it after I get back from Chicago. Here's a very informative video that explains in detail how to get the VCP200 set up: https://youtu.be/kFth9K_IvwA Here's the final project: Here's a test of the final prototype with Jet Set Willy Please don't criticize my gameplay as I was concentrating on the voice commands https://youtu.be/aov_Zx8oFAM 1 Quote Link to comment Share on other sites More sharing options...
Opry99er Posted October 30, 2015 Share Posted October 30, 2015 Man, that is cool!!! Quote Link to comment Share on other sites More sharing options...
+Vorticon Posted October 30, 2015 Author Share Posted October 30, 2015 Thanks In the end though, voice recognition on a retro computer is mostly a gimmick, but a fun one nonetheless! This has been one of my favorite hardware projects for the TI to date. 1 Quote Link to comment Share on other sites More sharing options...
+Vorticon Posted October 30, 2015 Author Share Posted October 30, 2015 Here are the design files if someone wants to reproduce the project: 1 Quote Link to comment Share on other sites More sharing options...
RickyDean Posted October 30, 2015 Share Posted October 30, 2015 Here are the design files if someone wants to reproduce the project: I actually think I have one of these in storage somewhere, I remember wanting to dabble with it in the 90's. Quote Link to comment Share on other sites More sharing options...
+Vorticon Posted October 30, 2015 Author Share Posted October 30, 2015 Yeah, that was the case with me, until I actually decided to finally do something with it a few decades later Quote Link to comment Share on other sites More sharing options...
Like us on Facebook
Follow us on Twitter
AtariAge on Instagram
Watch us on YouTube
Chat on Discord
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.