| PS3XG | Ag3 | m | (dr rob fletcher, age 40, computer support department administrator) unspecified |
| PS3XH | Ag2 | m | (aaron turner, age 26, computer technician, Speaks (fascinatingly) with both a stammer and a stutter.) unspecified |
| dr rob fletcher (PS3XG) |
[1] [...] need to see where you've got to and |
| aaron turner (PS3XH) |
[2] Well ... the software came off the tape without any any problems wha whatsoever. |
| dr rob fletcher (PS3XG) |
[3] Good. |
| aaron turner (PS3XH) |
[4] And I copied it over It all compiled first time. |
| dr rob fletcher (PS3XG) |
[5] E even the Pascal? |
| aaron turner (PS3XH) |
[6] Even the Pascal. |
| dr rob fletcher (PS3XG) |
[7] That's interesting cos I mean it is a sort of a [...] variant |
| aaron turner (PS3XH) |
[8] Luckily we have the er compiler for that. |
| dr rob fletcher (PS3XG) |
[9] We have that particular compiler, that was |
| aaron turner (PS3XH) |
[10] Yeah. |
| dr rob fletcher (PS3XG) |
[11] good then. |
| aaron turner (PS3XH) |
[12] Yeah. |
| dr rob fletcher (PS3XG) |
[13] I I thought you might in fact but I mean when it was first taken we first looked at the tape, people in there weren't sure so Oh that's good news oh that helps a lot because they've tried The original Moneytalk was developed on VAX V M S |
| aaron turner (PS3XH) |
[14] Right. |
| dr rob fletcher (PS3XG) |
[15] so that's why you see a lot of talk about VAX V M S around [...] mailboxes and funny things like that which are |
| aaron turner (PS3XH) |
[16] Yeah. |
| dr rob fletcher (PS3XG) |
[17] which are sort of er e Well a mailbox is a very crude inter process communication technique used on VAXes. [18] But I mean, I don't think you need to worry about that cos I think [...] got round that when they when they converted up to. [19] Should be okay. [20] Yeah so that all compiled up so what some bits don't work? |
| aaron turner (PS3XH) |
[21] Erm there we there was a problem with the erm lexicon [...] lexicon |
| dr rob fletcher (PS3XG) |
[22] Right. |
| aaron turner (PS3XH) |
[23] That just core dumped. [24] But I was able to transfer it onto a separate system and compile that and produce the lexicon there. |
| dr rob fletcher (PS3XG) |
[25] So why did that core dump on on [...] |
| aaron turner (PS3XH) |
[26] I have no idea. [27] I ... look looked through it, couldn't actually see anything obviously wrong with it. [28] I compiled it onto another system which has far better de debugging tools and |
| dr rob fletcher (PS3XG) |
[29] Yeah? |
| aaron turner (PS3XH) |
[30] It didn't core dump. |
| dr rob fletcher (PS3XG) |
[31] But that that that sudden? |
| aaron turner (PS3XH) |
[32] Well [...] |
| dr rob fletcher (PS3XG) | [...] |
| aaron turner (PS3XH) |
[33] Hewlett Hewlett Packard. |
| dr rob fletcher (PS3XG) |
[34] Oh you did one on the H P. [35] Which one was the one that worked? |
| aaron turner (PS3XH) |
[36] The one on the H P. |
| dr rob fletcher (PS3XG) |
[37] The H P worked better. |
| aaron turner (PS3XH) |
[38] Yeah. |
| dr rob fletcher (PS3XG) |
[39] Right. [40] But it doesn't wor do it doesn't [...] So it works on the H P but not on the Sun is that right? |
| aaron turner (PS3XH) |
[41] Yeah. [42] So I created the lexicon |
| dr rob fletcher (PS3XG) |
[43] Right okay. |
| aaron turner (PS3XH) |
[44] on the H the H P and copied that back |
| dr rob fletcher (PS3XG) |
[45] Yeah sure. |
| aaron turner (PS3XH) |
[46] and ... the first stage which is the for formatting stage |
| dr rob fletcher (PS3XG) |
[47] Mhm. |
| aaron turner (PS3XH) |
[48] works. |
| dr rob fletcher (PS3XG) |
[49] Right. |
| aaron turner (PS3XH) |
[50] As far as I can tell, the other stages also seem to work, |
| dr rob fletcher (PS3XG) |
[51] Right. |
| aaron turner (PS3XH) |
[52] but there is a problem with the the second stage which decomposes the |
| dr rob fletcher (PS3XG) |
[53] Right [...] |
| aaron turner (PS3XH) |
[54] formatted input. |
| dr rob fletcher (PS3XG) |
[55] Right. |
| aaron turner (PS3XH) |
[56] Er into parts of parts of speech. |
| dr rob fletcher (PS3XG) |
[57] Mhm. |
| aaron turner (PS3XH) |
[58] If I tried to give it more than one one word at a time it falls over and core dumps. |
| dr rob fletcher (PS3XG) |
[59] Right. |
| aaron turner (PS3XH) |
[60] With one word it's unable to determine what what part of speech it is. |
| dr rob fletcher (PS3XG) |
[61] This is not unreasonable. [laugh] |
| aaron turner (PS3XH) |
[62] Yeah. |
| dr rob fletcher (PS3XG) |
[63] Yeah. |
| aaron turner (PS3XH) |
[64] And ... I've been been through er th [...] the debugging tools. [65] [...] Pascal on on the main system. |
| dr rob fletcher (PS3XG) |
[66] Yeah. |
| aaron turner (PS3XH) |
[67] And slowly [...] up uphill struggle to actually [...] |
| dr rob fletcher (PS3XG) |
[68] Have you managed to locate it at al Not been able to locate where it actually |
| aaron turner (PS3XH) |
[69] Erm |
| dr rob fletcher (PS3XG) | [...] |
| aaron turner (PS3XH) |
[70] I've been able to trace back it's probably in a procedure called fetch page. [71] That seemed to be causing |
| dr rob fletcher (PS3XG) |
[72] Well is that fetching I mean is that a is that the routine that's picking information back from the lexicon or something ? |
| aaron turner (PS3XH) |
[73] Yeah. |
| dr rob fletcher (PS3XG) |
[74] Right. [75] I was wondering I mean ... there may possibly be differences between I mean there was [...] the format that speech generate of the H P and the format that the Sun would use. |
| aaron turner (PS3XH) |
[76] Mm. |
| dr rob fletcher (PS3XG) |
[77] Er now I would think that the lexicon itself is just a w a whacking great look up table yes? |
| aaron turner (PS3XH) |
[78] Yeah. |
| dr rob fletcher (PS3XG) |
[79] Right. [80] Er I would check things like, signed and unsigned bytes. [81] That's the that is the first problem you might find. [82] Have you checked that at all? |
| aaron turner (PS3XH) | [...] |
| dr rob fletcher (PS3XG) |
[83] So you might find that if you do a thing on the H P you may find that what is being what's happening is that it it it's creating ... the wrong size addressings in the look up table because either it's using you know, er signed and it's su assuming unsigned byte or something. [84] So you've got the data ranges incorrect, that sort of thing . |
| aaron turner (PS3XH) |
[85] Right. |
| dr rob fletcher (PS3XG) |
[86] Er you want to check that specification. [87] That's quite likely where it's going to be going wrong. |
| aaron turner (PS3XH) |
[88] Right. |
| dr rob fletcher (PS3XG) |
[89] That I I've seen that before between erm VAX systems and Silicon Graphics systems. [90] I also I mean Sun to Silicon Graphics, sometimes you need to make s see what the compiler option is in fact. [91] You may find the default compiler option for the H P is different to the default compiler option to handle signed and unsigned integers or signed and unsigned bytes, may be different. [92] It may well be the case. [93] Er now these are you know is the lexicon a Pascal module or a ... C module? |
| aaron turner (PS3XH) |
[94] It's a C module. |
| dr rob fletcher (PS3XG) |
[95] C right, so I would I I would almost would [...] think |
| aaron turner (PS3XH) |
[96] Right. |
| dr rob fletcher (PS3XG) |
[97] it's a sign problem in that it's not doing the look up properly. [98] These things are hideous to sort out I know. [99] Erm w I mean where they don't work when somebody offers code and they don't work, there's not [laughing] a lot you can do other than sort of make some informed guesses as to what might be going wrong. [100] Because at the end of the day I mean they're building up a a set of rules. [101] [...] practically you've got like a rule compiler. [102] I mean the the later stages of the system had things like rule compilers and quite complex things [...] which allowed you a more dynamic interaction with the system so you could actually build rules in and then [...] some rules then erm allowed you to ... shall we say, ... partition out certain parts of speech. [103] You found sometimes that other rules weren't required because that rule new rule that you'd finally developed encompassed other classes of events. [104] You know so so the idea was having a rule compiler system which so you could actively refine it. [105] You know the rules got more complex but in but in terms of actual processing power, the amount of look up required went down. [106] One of the reasons for big look up tables is that the [...] prototype system [...] . [107] Eventually the whole lot would be blown into ROMS. [108] In a box. [109] [laugh] . Well [...] in fact it was [...] . [110] And still can ma buy cards for a P C [...] . [111] But I'm not sure whether that would be in of interest at all to Ian anyway. |
| aaron turner (PS3XH) |
[112] That might might |
| dr rob fletcher (PS3XG) |
[113] Yes you can actually buy a [...] card for a P C. And all you do is you feed it erm text and it'll speak it. |
| aaron turner (PS3XH) |
[114] Yeah. |
| dr rob fletcher (PS3XG) |
[115] Albeit in an American voice. |
| aaron turner (PS3XH) |
[116] Is erm Ian definitely wants wants new suns. |
| dr rob fletcher (PS3XG) |
[117] Yes I know. |
| aaron turner (PS3XH) |
[118] [...] being targeted at students [...] . |
| dr rob fletcher (PS3XG) |
[119] Yeah I mean the [...] would be very nice if you could get it to go cos it is it is quite acceptable speech. [120] Albeit American. [121] [laugh] American English. [122] Which is different of course. [123] So you don't get dog, you get dog. [124] [laugh] . [...] D A W G. [laugh] . [125] Er but I think people can live with that [...] understand sort of Americanisms now. [126] [...] Erm that that's I think the main thing you want to look at there. |
| aaron turner (PS3XH) |
[127] Right. |
| dr rob fletcher (PS3XG) |
[128] Er what about the synthesis stage, do you know whether that works or not? [129] Does it generate waveforms? |
| aaron turner (PS3XH) |
[130] It's been a bit Well it's it's been been a bit difficult to test because with the sort of ready the read-me files [...] . [131] Very very minimal so I haven't noticed [...] |
| dr rob fletcher (PS3XG) |
[132] Were there any erm already er sort of the intermediate files? [133] Things like er erm s [...] synthesizer files. |
| aaron turner (PS3XH) |
[134] Nothing. |
| dr rob fletcher (PS3XG) |
[135] [...] parameter file. [136] Er just wondering whether |
| aaron turner (PS3XH) |
[137] There is apparen apparently book books details of [...] of course it's out at the moment. |
| dr rob fletcher (PS3XG) |
[138] Mm. |
| aaron turner (PS3XH) |
[139] I could search round for papers on it. |
| dr rob fletcher (PS3XG) |
[140] There's not there's not a lot on because n Well I mean,w the reason why there's not a huge amount published is because er Digital bought the rights to DECTalk. |
| aaron turner (PS3XH) |
[141] Right. |
| dr rob fletcher (PS3XG) |
[142] For something like six million dollars. [143] And so at that point, nothing very much was published. [144] No not in the public domain anyway. [145] I mean there are plenty of internal reports at M I T and at DEC but you certainly can't get them [...] . [146] I have seen a paper [...] been involved with M I T themselves. [147] [...] Er do you know what the er what a the input file would look like to the synthesizer phase? |
| aaron turner (PS3XH) |
[148] It gives a very sketchy idea in the read-me files however it also |
| dr rob fletcher (PS3XG) |
[149] Do you know how many control parameters there are for instance? [150] Is it the twenty one or is it the sixty or forty eight [...] |
| aaron turner (PS3XH) |
[151] Without looking through the source file I wouldn't be able to tell. |
| dr rob fletcher (PS3XG) |
[152] Because erm I do have a variation er a version of the that particular synthesizer [...] which is a class seventy nine synthesizer. [153] And I may well have some input files which you could u test with. [154] You need to find out though what version that that is. [155] It may well be well the one that was out in the public in the public domain. [156] One that was published in ... the Journal the American Society for Acoustical Research. [157] J A S A, J A S A. |
| aaron turner (PS3XH) |
[158] J S? |
| dr rob fletcher (PS3XG) |
[159] J A S A. Journal of the American Acoustical Society. [160] ... Journal of the American Society for Acoustics, something like that anyway. [161] That that that the original class seventy nine synthesizer was published, the source code [...] . [162] And it was originally in FORTRAN source so I don't know whether [...] is it still in FORTRAN source? |
| aaron turner (PS3XH) |
[163] Erm no it's Pascal now. |
| dr rob fletcher (PS3XG) |
[164] Is it Pascal source now. [165] Yeah n what you need to do is if you can look up the what the input parameters h how many input parameters there are, and if there are ... if if it's basically a batch of twenty one parameters, er I can probably I've probably got the source code for them. [166] Er I mean the input files. [167] It depends whether they're ASCII input or whether they're binary input format. [168] I would put money on them being what are called K P R |
| aaron turner (PS3XH) |
[169] Yeah. |
| dr rob fletcher (PS3XG) |
[170] Er K P R format which is [...] parameter file. [171] Er P R for parameter. [172] [laugh] . [...] parameter file and that is actually effectively a a sixteen bit integer encoded file. [173] And in fact I can probably get you some of those as well. |
| aaron turner (PS3XH) |
[174] Right. |
| dr rob fletcher (PS3XG) |
[175] Okay. [176] Er we moved over to using ASCII so that we could actually hand edit them [laugh] and and look at them. [177] But this is about four or five years ago so I mean it's n we've moved on a lot since then. [178] But certainly we can probably find some stuff for [...] er to help test. [179] But did you get er but does the pipeline build for instance? [180] I mean there's a program that builds the pipeline. |
| aaron turner (PS3XH) |
[181] Yeah. |
| dr rob fletcher (PS3XG) |
[182] And that all builds? [183] Right well that's wonderful, I mean that's really good news really, |
| aaron turner (PS3XH) |
[184] Mm. [185] Mm. |
| dr rob fletcher (PS3XG) |
[186] You see cos a again it was probably in the early days, running on P D P Elevens. |
| aaron turner (PS3XH) |
[187] Right. |
| dr rob fletcher (PS3XG) |
[188] Right so you've got to think P D P Eleven because it's fairly old stuff now. [189] And I mean Unix in those d in the early days was P D P Eleven based. [190] And then it moved over to sort of VAXes as well. [191] Er but I think [...] could be could be the backspace [...] . [192] Cos it's using B er B S D isn't it or [...] J S D stuff so. [193] Could be could well could have been backspace but still there'd still be Eleven stuff around I would have thought. [194] Er you'd probably find that in the [...] P D P Eleven around the place. [195] [laugh] . [...] . And if it is then they're probably sixteen bit integer encoded files for the synthesizer phase. [196] The other thing is that the output frequency is likely to be eight kilohertz or ten kilohertz. [197] Er you need to think about that. [198] But again, depending on your hardware you may the output phase hardware, you might need to alter some things in the synthesizer to change it to the from eight to ten to twelve to sixteen bit data. [199] Again I I modified the seventy nine and rewritten the seventy nine [...] I've had also the the later version of the synthesizer to handle different bit-width so it's not a problem. [200] And I know how to do that. [201] There's not a lot about to change. [202] You know you tend to find there are a couple of fiddle factors around actually [laugh] . [203] And magic numbers like multiplying the width of the byte by sixteen point O two five or something [laugh] . [204] [...] . You m is it really well is it very well commented at all? |
| aaron turner (PS3XH) |
[205] It's fairly well commented. |
| dr rob fletcher (PS3XG) |
[206] Yeah cos I I think you'll find there are various comments like you know, let's say you know, move the the data into the highest signi er the most significant bits of the word and this sort of stuff. [207] You know where they and they just multiplying by number rather than doing bit shifts [laugh] . [208] So [...] So I mean you'll you'll find I mean, it's not that bad. [209] Right I well if you can't get the lexicon going it's got a problem of course. [210] [...] . If it was on a P D P Eleven then ... the addressing would only be a matter of si [...] sixteen bit as well. [211] In the tables. [212] So again you might find that that you're generating the wrong type of integer number. [213] I mean between your two machines. [214] And also possibly with what it puts as a co Was it expect You know have a read of the code |
| aaron turner (PS3XH) |
[215] Yeah. |
| dr rob fletcher (PS3XG) |
[216] a little bit more of the code, just to find out you know, what size integers it's using. [217] And I think that's probably where your pro the problems would be . |
| aaron turner (PS3XH) |
[218] Mm yes yes. |
| dr rob fletcher (PS3XG) |
[219] Mm [...] sure. |
| aaron turner (PS3XH) |
[220] That sounds likely. |
| dr rob fletcher (PS3XG) |
[221] You know I've seen those sorts of problems before. [222] [...] Trying to think what else there is. [223] Er that's about what there is well there's nothing to look at. [224] At this stage anyway, then we need to have a another think. [225] A little bit of detective work I think [laugh] to try and find out. [226] I mean you could even write some fairly simple test programs in fact. |
| aaron turner (PS3XH) |
[227] Mm. |
| dr rob fletcher (PS3XG) |
[228] That do a little bit of you know, very simple table look up [...] . [229] Or build a table. [230] Does the l does the lexicon generate you the C source code? [231] Like a a C look up table? [232] Or does it generate a file? |
| aaron turner (PS3XH) |
[233] It generates a large well |
| dr rob fletcher (PS3XG) |
[234] A large C large C structure. [235] Or a just a large table or a large array. |
| aaron turner (PS3XH) |
[236] Erm ... g erm it just gen gen generates a large large [...] . |
| dr rob fletcher (PS3XG) |
[237] You see cos I know that some of the phases you find that ... Does it genera you see cos some some of the programs generate programs. [238] [laugh] In the in the later versions certainly. [239] But you you know you [...] generate some of the C source or assembler source which you then compile. [240] No I assume that your file is not one of those. |
| aaron turner (PS3XH) |
[241] No. |
| dr rob fletcher (PS3XG) |
[242] No okay. [243] S Yeah I wo I would basically put money on it being a problem with the the addressing. [244] And it's gonna be either what a a a short it is or what a [...] . [245] [...] [laugh] . Yeah I I if there's anything else you want to know I mean I think that's quite a good start [...] |
| aaron turner (PS3XH) |
[246] Erm yeah I'll er |
| dr rob fletcher (PS3XG) |
[247] one of those problems anyway. |
| aaron turner (PS3XH) |
[248] I'll go through that [...] |
| dr rob fletcher (PS3XG) |
[249] I I I would think [...] I can't really see it being much else, I mean it compiles up. [250] That's what it's most likely to be I think. [251] [laugh] Okay. [252] If there's anything else you want to know? [laugh] [...] |
| aaron turner (PS3XH) |
[253] [...] . |
| dr rob fletcher (PS3XG) |
[254] I'm generally around so you know. [255] Give me a shout. [256] Don't lose your bit of paper. |
| aaron turner (PS3XH) |
[257] Okay. |
| dr rob fletcher (PS3XG) |
[258] Yeah that's fine. |
| aaron turner (PS3XH) |
[259] Thank you very much. |
| dr rob fletcher (PS3XG) |
[260] It's no problem. [recording ends] |