For now mkspeech only supports mp3, but it is VERY easy to add other formats...
I think the remaining major task is to choose the correct line when the result is ambiguous.
See the missing.txt that is created for reference on what is not unique.
Another little update at the same URL. Now quotes are translated as well as \xFE\x08 is ignored.
Coverage is now up to 3946 lines!
There are not many real strings left now.
I haven't used cocomonk22 modified version which actually should increase the coverage even more as those lines with variables are separated.
Is it available somewhere? Or are there some detailed instructions which scripts to modify etc.? (Especially with redistribution in mind...)
I haven't used cocomonk22 modified version which actually should increase the coverage even more as those lines with variables are separated.
Is it available somewhere? Or are there some detailed instructions which scripts to modify etc.? (Especially with redistribution in mind...)
Check your PMs.
I haven't used cocomonk22 modified version which actually should increase the coverage even more as those lines with variables are separated.
Is it available somewhere? Or are there some detailed instructions which scripts to modify etc.? (Especially with redistribution in mind...)Well. There is the speech project talkie patch. I guess, you can use that, if you remove all speech references from it before using the tools here.
Well. There is the speech project talkie patch. I guess, you can use that, if you remove all speech references from it before using the tools here.
That's exactly what cocomonk22 sent me; a hack-only version of the MISP patcher.
Another update :D....
Now it's up to 4046 matches! Now there is really not much left!
Time to add entries to the mapping.txt!
Without variables in the string table (with cocomonk22 et al. changes) it's now 4436 of 4573 available samples assigned.
I would call this a pretty neat coverage :)
The only real problem now are the false positive matches that are listed in the missing.txt....
I'm missing speech on any line with a TM symbol. This is looking (sounding?) better and better though!
TM is working now, that makes it 4511 of 4573.
As said it would be good if there would be some volunteers that add the correct lines in mapping.txt. This is something *I* really don't want to do .....
If there is more than one volunteer, the work should be split up. Like line 1-2200 for person one and 2200-end for person two.
EDIT: You may want to edit the first posting, to reflect the current state of affairs.
As said it would be good if there would be some volunteers that add the correct lines in mapping.txt. This is something *I* really don't want to do .....
I don't really understand what's to do here...
I've ran mkspeech.exe with this result:
Read 4573 entries.
Could not open mapping.txt
Could not find SLP_71_gh-room_9_1.mp3
Wrote 4121 samples, 1194 missing.
What I got is an mi1new.txt, a missing.txt and the monster.so3
What now?
What I got is an mi1new.txt, a missing.txt and the monster.so3
What now?
Next step is to integrate the mi1new.txt into the original game files (make sure you have a backup...):
scummtr -cw -g monkeycdalt -if mi1new.txt -H
When putting the monster.so3 in the MI1 game folder and you run it via ScummVM you should hear voices (just to make sure everything works for you).
You will notice right from the beginning that there are some errors as "Hi!" as a string occurs multiple of times and thus there is no way(*) to automatically detect which "Hi!" is used.
When you look at the "missing.txt" you should see something like:
Multiple hits for line "Hi!": HRM_80_fort_25_1.mp3 - Add 584,0=975 to mapping.txt
Multiple hits for line "Hi!": GUY_38_lookout_28_1.mp3 - Add 584,0=2306 to mapping.txt
Multiple hits for line "Hi!": HRM_20_main-beach_48_1.mp3 - Add 584,0=2604 to mapping.txt
Multiple hits for line "Hi!": GUY_20_main-beach_58_1.mp3 - Add 584,0=4134 to mapping.txt
Probably four times.
So it's about finding out, which "Hi!" is the correct in all the four places.
When adding "584,0=4134" to mapping.txt, the next call of "mkspeech" should pick the last sample for that particular line (not sure if this is the right association - this is just an example).
Finally you need to call scummtr again, to integrate the new text file (and make sure to use the new moster.so3 too).
P.S.:
I think a little problem occurs depending on the version we use as the patches from cocomonk22 change the order of strings a bit.
I haven't put any thought on how to make it a bit more robust in that regard....
But if the version is defined properly for the file, it will not be hard to convert it later...
(*) There might be a way, but it would probably involve modifying (or rebuilding) scummtr to get the room ids. Unfortunately I have not found the source code for scummtr so this does not seem to be an option..... well having scummtr source would also allow to build a single application that performs ALL steps :-(
Would it be possible to offer this thing as a "patch"? That way only people who own the SE will be able to pull this off, and there's no piracy involved (which would occur when sharing the actual speech files). Maybe a batch-file that pulls the speech files out of the SE's PAK, converts them accordingly (and possibly combines split lines into a single file) and re-inserts them into the CD-version of the game using ScummSpeaks.
This would allow to share this project without running into legal trouble.
that would be great if you could do that. It would help people (Especialy Me) to not have to buy a new one.
that would be great if you could do that. It would help people (Especialy Me) to not have to buy a new one.
Yes, we're going to offer this as a patch... but what do you mean "not have to buy a new one"?
that would be great if you could do that. It would help people (Especialy Me) to not have to buy a new one.
Well actually that's the aim. For now the only thing needed is MISE, as it contains the original enhanced CD version.
With my both tools unxwb-mi and mkspeech (maybe I'll integrate the relevant unxwb part in mkspeech at a later point), it's already possible to get a basic speech version of the classic enhanced MI version. The mapping file also is unproblematic to redistribute.
And finally a batch file that is being worked on by Espiox will make it hopefully relatively easy to generate your own version without many manual steps.
I think the only problem are the patches by cocomonk22, where I am not sure on what their legal status is and if they can be redistributed freely.
Why can't the patches I made be redistributed freely?
I get a problem when I disable the subtitles: Those multi-part lines don't work anymore... only the first part is played.
Also some lines overlap (for example when talking to the one-eyed pirate in the Scumm Bar).
Edit:
And sorry, I still don't get how to put together a "mapping.txt". This is how the first few lines of my missing.txt look like:
Missing "jungle", add 1,0=sampleno
Missing "rowboat and oars", add 2,0=sampleno
Missing "jungle", add 3,0=sampleno
Missing "jungle", add 4,0=sampleno
Missing "jungle", add 5,0=sampleno
Multiple hits for line "It's a gross old bottle.": GUY_1_beach_1_1.mp3 - Add 6,0=228 to mapping.txt
Multiple hits for line "It's a gross old bottle.": GUY_1_beach_1_1.mp3 - Add 6,0=230 to mapping.txt
Multiple hits for line "I don't want it.": GUY_1_beach_1_2.mp3 - Add 6,1=229 to mapping.txt
Multiple hits for line "I don't want it.": GUY_1_beach_1_2.mp3 - Add 6,1=231 to mapping.txt
Multiple hits for line "I don't want it.": GUY_30_store_38_2.mp3 - Add 6,1=3923 to mapping.txt
Multiple hits for line "I don't want it.": GUY_30_store_58_1.mp3 - Add 6,1=3953 to mapping.txt
Multiple hits for line "I don't want it.": GUY_30_store_62_1.mp3 - Add 6,1=3964 to mapping.txt
Missing "bottle", add 7,0=sampleno
Missing "note", add 8,0=sampleno
Missing "Herman Toothrot", add 9,0=sampleno
Missing "beach", add 13,0=sampleno
Missing "beach", add 14,0=sampleno
Missing "crack", add 15,0=sampleno
So, the first 5 entries are probably lines, that aren't spoken but only text. So I don't add them to the mapping.txt, because they don't need to be associated with any audio-files, right?
The 6th entry is looking for the correct "It's a gross old bottle"-audio, but since those are obviously linked to the same audio-file, I just pick on of them and add 6,0=228 to the mapping.txt. Entry 6,1 (I guess ",1" is the response when trying to "pick it up") is looking for the correct "I don't want it", which should be a little easier, since I know it's said by Guybrush on the beach. So I'll add 6,1=229 to the mapping.txt, since that's the correct Guybrush line.
But what if I don't know where that line occurs? How am I supposed to know, which sample is the correct one? (Okay, sometimes the surrounding lines give clues about the place, as do the name of the audio-files)
And, do I only have to add stuff to the mappings.txt that's listed as "multiple hits" in the missing.txt?
Why can't the patches I made be redistributed freely?
Oh I did not say that. I just wondered if there is an "official" place for them as it seems to be a bit... hidden... please if there are no objections, we should put all relevant links in the first posting.
I get a problem when I disable the subtitles: Those multi-part lines don't work anymore... only the first part is played.
Hmm that's bad... I am not sure if there is a way to fix it.
I guess we could either go back to merging the samples (which would bring the text and speech out of sync....), look if ScummVM would accept a patch that fixes this behavior or modify the game code so that only single lines are used.... (or find out if the tag field might be useful here too)
Also some lines overlap (for example when talking to the one-eyed pirate in the Scumm Bar).
Yes, I noticed that too. Not sure what to do here either. Maybe the "tags" field can be used to prevent this?!
So, the first 5 entries are probably lines, that aren't spoken but only text. So I don't add them to the mapping.txt, because they don't need to be associated with any audio-files, right?
Right. Those can be ignored.
But what if I don't know where that line occurs? How am I supposed to know, which sample is the correct one? (Okay, sometimes the surrounding lines give clues about the place, as do the name of the audio-files)
First: The second number ".,X=..." is used to determine the section when there are multiple lines in one row. Like "Hi!" "My name is Guybrush Threepwood!"
And yes, the main task is to find out where a line is said and which is the correct one. For some the context might give a hint, for other there is probably no other way but playing the game to the point where it is said.
And, do I only have to add stuff to the mappings.txt that's listed as "multiple hits" in the missing.txt?
In theory, yes. There might be fields with a missing sample where the proper sample exists and has to be determined.... but I think it should not happen that often.
I would say the priority should be to find the right matches for the ambiguous lines, as those are also the ones which will sound *wrong*. When there is no sample played it's not as strange as when the wrong one is played ....
I think there's more problems... even with the subtitles on, the timing in some scenes is completely off, for example with the Fettucini Brothers.
I think there's more problems... even with the subtitles on, the timing in some scenes is completely off, for example with the Fettucini Brothers.
Hmm strange, this particular scene seems to work fine here.
I'm using the latest SVN build of ScummVM btw....
Anyway, if somebody has some inside knowledge of the exact tag definition that (s)he is willing to share, we could probably fix some of the problems....
Laserschwert, the problem's probably on your end. Are you using my patched version, latest ScummVM build, etc?
I'm using the latest SVN as well, and the Fettucini's keep cutting each other's lines off. (or overlap)
But what patched version are you talking about? Obviously I've missed that...
Thanks, the Fettucini-part works now. But there's still the problem of multi-part lines not working with the subtitles turned off, and there are still overlaps (for example when talking to the lookout).
Hmm strange, this particular scene seems to work fine here.
I'm using the latest SVN build of ScummVM btw....
Anyway, if somebody has some inside knowledge of the exact tag definition that (s)he is willing to share, we could probably fix some of the problems....
Probably not what you're asking, but I know how lip-synch tags are stored in SCUMM V5.
Warning! Techno-babble ahead
Tags are 16-bit words stored in big-endian format.
Text:
Ignoring the \xFF\x0A bits, your first four values are four bytes for the speech offset (little-endian), and the last four values are four bytes (little-endian) representing the number of lip-synch tags as modified by this formula:
(numtags << 1) + 8
e.g. To play a sample at offset 0x1234 with 2 lip-synch tags:
\xFF\x0A\x34\x12\xFF\x0A\x00\x00\xFF\x0A\x0C\x00\x FF\x0A\x00\x00
Compressed MONSTER.SOU:
The number of tags are stored for each sound, as part of the table at the start of the file. The tags themselves are written before the MP3/OGG/FLAC data (e.g. for a sound at 0x1234 with two tags, 0x1234 to 0x1238 stores the tags, and 0x1238 onwards stores the sound data).
Tag format:
Each tag is just a time position in the sound file being played (I'm not sure what measurement the position is in, maybe milliseconds, maybe not). Whenever the next tag's position is encountered, the talking animation is toggled on or off.
e.g. if there are three tags, with values 120, 240, 640, the animation timeline will play like this:
0-120: talking
120-240: not talking
240-640: talking
640-end: not talking
I think one problem is that the old SCUMM games did not support multiple speech sounds per line. Check Indiana Jones and the Fate of Atlantis, and I'm sure you'll find that all multi-line speech just uses the one sound file, and the subtitles are frequently out of synch with the sound.
Sorry I'm late to the party!
Check Indiana Jones and the Fate of Atlantis, and I'm sure you'll find that all multi-line speech just uses the one sound file, and the subtitles are frequently out of synch with the sound.As I did checked above. Thus, we should implement it the same way. If we can't split the lines in individual talking events, we have to join the samples.
Do all subtitle lines have a constant display duration?
Or is there a way to modify this duration per line, possibly through a modified version of scummtr, so that it would match the length of the corresponding speech sound?
Do all subtitle lines have a constant display duration?
I believe the subtitle display duration is calculated based on the length of the subtitle, modified by the global subtitle speed.
(numtags << 1) + 8
Ah! That's exactly what I was missing. Thanks.
Whenever the next tag's position is encountered, the talking animation is toggled on or off.
Ok, this will not help with multiple lines. I wonder if it helps with cut off samples when having the last tag >= length of the sample...
I think one problem is that the old SCUMM games did not support multiple speech sounds per line. Check Indiana Jones and the Fate of Atlantis, and I'm sure you'll find that all multi-line speech just uses the one sound file, and the subtitles are frequently out of synch with the sound.
While that is true, I think we can expect people to use ScummVM when they want a proper Talkie version. And I was somehow hoping we could find a trick this way...
As said, multiple lines are in fact working fine with ScummVM only that they are messed up when the subtitles are turned off.
Well that said, the tags are definitely no solution for this problem, as they are just for lip-animation timing nothing more.
So this finally leaves us this four options:
1) Try to patch ScummVM and bring the patch upstream.
2) Modify the scripts so no more multi-lines are used.
3) Merge audio files and have out of sync subtitles.
4) Force the user to always have subtitles enabled.
Or am I missing something?
IMHO 1) or 2) are the better solutions as they should not have any negative side-effects.
Sorry I'm late to the party!
Hey, the party just started :)
I believe the subtitle display duration is calculated based on the length of the subtitle, modified by the global subtitle speed.There are occasions in which they are cut short, though. For instance, the introduction to the Fettucini brothers is supposed to be fast talk, Stans lines about personal problems too. And the pirate leaders "grog grog grog" and "ha ha ha" are also short for the flashing effect. There are probably more instances.As said, multiple lines are in fact working fine with ScummVM only that they are messed up when the subtitles are turned off.Couldn't we just submit a patch for ScummVM? Eventhough, it is no intended original LEC feature, it won't hurt to have this supported, I'd say.
As a side effect, this would also improve subtitle sync with FOA etc., since those have indeed the same issue.
The only real problem now are the false positive matches that are listed in the missing.txt....I think, most of them can be avoided by checking the room number. Passing the -h parameter to scummtr adds a leading info from where the line comes, starting with the room number. Conveniently, most samples are also named with the room number.
Most ambiguous matches would be avoided that way. Only a few remain in cases where the same line is used in the same room, like the cannibals' "nothing.". And I guess, all the insults, too.
Couldn't we just submit a patch for ScummVM? Eventhough, it is no intended original LEC feature, it won't hurt to have this supported, I'd say.
As a side effect, this would also improve subtitle sync with FOA etc., since those have indeed the same issue.
Well we could, but it will not improve FOA with the current approach as FOA only uses one sample per line. (Well a mod could probably be made .. but I wonder if there would be any people willing to do so....).
Fell free to make a patch though :-)
I think, most of them can be avoided by checking the room number. Passing the -h parameter to scummtr adds a leading info from where the line comes, starting with the room number. Conveniently, most samples are also named with the room number.
Oh! I must have overlooked the -h option. Well yes, this could indeed improve it a lot.
Ok, with the hint of LogicDeLuxe it works MUCH better now.
EDIT: It also matches the interaction/script number now. So even less collisions.
The input file needs to be generated with:
scummtr -cw -g monkeycdalt -of mi1.txt -h -H
For the output it remains the same (the meta-info is not copied):
scummtr -cw -g monkeycdalt -if mi1new.txt -H
As usual the updated version at the same place:
http://helicoid.de/scumm/mkspeech.zip)
P.S.:
Stupid question: How to disable the subtitles for MI1 in ScummVM? I have set it to "Speak only" but the subtitles still pop up...
P.P.S.:
Well the room number for Spiffy is wrong ..... hardcoded the change.
I also filter all "OBNA" lines now....
Stupid question: How to disable the subtitles for MI1 in ScummVM? I have set it to "Speak only" but the subtitles still pop up...
You set it in the launcher, right? Setting it there doesn't do anything for me either. But when selecting it in the game it works.
I'm aiming to get a working version of my installer up tonight to make it easier for people to join in and test the speech.
EDIT: I'm having a bit of trouble getting it to the run the encoder. I'll get this up ASAP, but it won't be tonight.
Well we could, but it will not improve FOA with the current approach as FOA only uses one sample per line.Well, I wrote that with single samples for multi-lines in mind. Thus what FOA uses as well. Depending on the subtitle speed used, it can indeed go quite a lot out of sync at those lines, too.
With joint samples, a short silence should be added at the joint. About 300 ms sounds fine to me in most situations. The voodoo lady would probably need a bit more when she is guessing Guybrush' name.Well the room number for Spiffy is wrong ..... hardcoded the change.I guess, you could first find all matches according to room numbers, and after that, ignoring the room numbers for the remaining lines. Hopefully, those aren't that many.
Also, the tweak file should be able to exclude samples. As a curiosity, there are some lines recorded, which actually are never spoken. For instance, you can talk to your crew and have the option "Keelhaul: to haul under the keel of a ship as punishment or torture.", which has a sample, but Guybrush actually refuses to say this. Also the snoring sample is huge, and unnecessary, as there is already a much smaller sample for this in the game data file. (Unless, we want to patch that out in turn, that is)
And while we have patched scripts already, is there any chance we could fix some bugs which were introduced to the enhaced CD, but not present in the VGA floppies? Like:
- Smoke at Smirk's close up. Look's like the same graphic as the steam from the voodoo cauldron, and the smoke in the catacombs.
- LeChuck's shaking the grog machine animation.
- Consistent text color of the priest.
- White background at the sleeping dogs message. (This one was already fixed in the German patch:
http://la-patches.gandimann.de/)
- Brown floor in the kitchen door in the Scumm bar (Also this one was already fixed in the German patch:
http://la-patches.gandimann.de/)
- Navigator's directions. He rarely seems to look to the front or back, even when those are the only ways to go.
- Picture breaking in the mansion uses EGA graphic for some reason.
- And if we ask Dominic Amato politely, maybe the stump joke and the Charles Atlas joke too.
I would take a look into those issues, but I don't know how to use descumm and some script compiler for this. Are there any guides for this?
Is it only me, or are there some audio-files extracted that are empty? Edit: Okay, it MIGHT be just a Winamp-issue with short samples.
Well, I wrote that with single samples for multi-lines in mind.
Well still I assume fixing FOA sync is not trivial (otherwise it would probably be done already). I also think that the current approach is better, as the lines are always in sync this way. And I bet it won't be that hard to fix the problem when subtitles are disabled.
Also, the tweak file should be able to exclude samples.
Yeah, I'll probably add something later. "x,y=-1" should do the job I guess..
As a curiosity, there are some lines recorded, which actually are never spoken.
I really like the Guybrush_UnknownFilename_07 sample *g*
Also the snoring sample is huge, and unnecessary, as there is already a much smaller sample for this in the game data file. (Unless, we want to patch that out in turn, that is)
Is it actually included at the moment? Only samples that have a matching text line are copied to the monster.soX anyway...
I would take a look into those issues, but I don't know how to use descumm and some script compiler for this. Are there any guides for this?
Feel free to do so! I never played around with descumm so I can't help you there.
I guess, you could first find all matches according to room numbers, and after that, ignoring the room numbers for the remaining lines. Hopefully, those aren't that many?
Well now it only matches if the room number is correct. I think there are really few rooms that do not match, so it's probably not worth the effort to add extra logic there.
The remaining lines are mostly the sword fight scenes.
Is it actually included at the moment? Only samples that have a matching text line are copied to the monster.soX anyway...I didn't check yet. The one at the brig has the snoring subtitled, so I assumed it is.
After some testing, I see that including the remaining sounds which didn't match a room would not only useful for the new Spiffy room (which seems to work, btw.), but also those which had variables in it. Those are currently silent.
I'm trying to use the new mkspeech tool, but when I put the mi1new.txt file back into the game, scummtr says: ERROR: NULL char in line 0. What am I doing wrong?
When adding the -I flag to scummtr.exe when creating the mi1.txt-file, it adds a hex-value to each line... what is this? Does it help mapping the files?
-I adds the opcode that initiates the text. For instance, D8 is used for printEgo (Guybrush talking), FA is for dialogue options, and 14 is for print (everyone else talking).
Doesn't this already help pulling Guybrush's lines out of the swordfighting lines?
Not really, as Guybrush's insults are in the normal place while I moved the pirates lines to the last room. Guybrush's insults have opcode 27.
Ok, I updated the tool (
http://helicoid.de/scumm/mkspeech.zip) again.
Now it should fall back to the next best line when the room does not match...
I'm not sure how much more time I will/can put into this...... well.... the source code is as always included. And I cleaned the code up at least a little bit.
Also a value of -2 in the mapping.txt should ignore the sample for the line... i.e. "66,1=-2". But this is untested.
I think the ultimate batch-file would contain a few questions first, including music replacement yes/no, narrator inclusion yes/no, Spiffy inclusion yes/no and so forth (I don't know if that last one is happening though).
And even better than having it just as a batch file would be a nice GUI for it... although of course I've got no idea how to do that.
I included my Spiffy addition with the patches I sent to Espiox. That's what SCUMM Image Encoder was for. I also had to create a new room for the Spiffy closeup scene.
http://www.youtube.com/watch?v=k6K8Ckzg9bI)
This is all such great news! I'm looking forward to it!
I've written a dialog line splitter: (outdated version)
It's not throughoutly tested yet. And there are at least these known bugs:
- Concerning 000_LECF\030_LFLF_store\000_ROOM\050_LSCR.dmp and 000_LECF\083_LFLF_cu-dock\000_ROOM\032_LSCR.dmp as mentioned here:
http://www.lucasforums.com/showthread.php?p=2654259#post2654259)
- scummtr reports some bad jumps.
You need to run scummpacker first. Then you need descumm, scummbler and this zip extracted in the same dir. Run dialogsplitter.bat and wait.
I think splitting the dialogue is the way to go, as that will enable the game to work properly with subtitles off.
EDIT: Jestar Jokin updated Scummbler, so now there are now errors on extracting the text with scummtr! We just need to get the audio matched with the right lines.
EDIT2: The dialogue splitter messed up Spiffy's text position after his first line, so I'm making a new patch now.
EDIT2: The dialogue splitter messed up Spiffy's text position after his first line, so I'm making a new patch now.Not only Spiffy's. I have to check what causes it. You better not working with the split patch, until this issue is solved.