Generating Chinese TTS

[Youlun]

Hi all, I'm trying to get Chinese TTS to work on my iPod video 5th generation.

I have the Microsoft Lili SAPI voice installed through the W7 Ultimate Simplified Chinese language pack (I tried the Traditional Chinese language pack, but it appears that Microsoft Lili only shows up after installing the simplified one). I now see something like this in Windows Speech Properties menu. In the Rockbox Utility under 'Configure -> TTS & Encoder', I've set the SAPI TTS Engine, and under 'Configure TTS' I've set the language to chinese-simp and the voice to Microsoft Lili (chinese-trad doesn't seem to have any voice options).

Now when I choose 'Create Voice File" in the Rockbox Utility the following happens:
chinese-trad: Finishes, but most menu items are voiced in English, using Microsoft Lili's voice.
chinese-simp: Fails with error - Encoding of LANG_BOOKMARK_DONT_RESUME failed

Here's the log from my chinese-simp voice creation attempt.

Thanks in advance for any help!

[bluebrother]

Is that TTS available for non-chinese systems? From a first look it seems to be a problem voicing an entry, and that could be related to codepage conversions.

[Youlun]

The TTS is available on non-Chinese systems; see here: http://mdbg.loqu8.com/tts.htm

Looking at the log, I'm suspicious of question marks I'm seeing instead of characters. For example:

Code: [Select]

[TalkGenerator] voicing:  "?取中 ?稍候" to C:/Users/.....

So perhaps you're right about it being a codepoint issue, though I'll admit that I don't know very much about these things.

[bluebrother]

So perhaps you're right about it being a codepoint issue, though I'll admit that I don't know very much about these things.

I've managed to reproduce this issue. Building a voice file using MS Lily works fine if I generate a voice file for english, and I also get the failure when using MS Anna for creating a voice for chinese-simp -- which makes me believe that it actually *is* an encoding related issue.

[Youlun]

Cool, glad to know I'm not the only one who's crazy. Should I make a bug report somewhere about this then?

[bluebrother]

Cool, glad to know I'm not the only one who's crazy. Should I make a bug report somewhere about this then?

Not necessary anymore (hopefully)

Short version: please try a current Rockbox Utility binary from http://tinyurl.com/rbutil-dev if you can successfully generate a chinese voice file. I was able to do so just minutes ago.

Long version:
It actually wasn't an encoding issue. After digging through this problem for two evenings it turned out that the chinese language file contains two voice strings that contain the characters < and > -- strings that are displayed with them (like "<All tracks>"). Obviously those characters don't make sense at all when voicing, so most language files have them removed from the voice string -- all translations that didn't were affected by this problem (basque, chinese-simp, russian). As a fix those characters are now simply removed before speaking the string. This works fine for me so unless you still have problems I consider that issue fixed

Edit: forgot to mention, the question marks in the log are caused by the logger trying to convert the characters to a local encoding. This doesn't affect passing the (correct) strings to the TTS, so it's not related -- but making the logger able to deal with extended characters would be a good thing

[Youlun]

Yup, chinese-simp is working with that build! I'm still seeing the problem with chinese-trad, where menu items are voiced by Microsoft Lili in English though. Is chinese-trad working for you?

[bluebrother]

Yup, chinese-simp is working with that build! I'm still seeing the problem with chinese-trad, where menu items are voiced by Microsoft Lili in English though. Is chinese-trad working for you?

Unless I'm getting you wrong here you're now talking about a different issue: voicefile generation works fine for both chinese-simp and chinese-trad, but chinese-trad containing english strings instead of chinese ones. Is that correct?

This would be a different issue -- for chinese-trad there seem to be a lot of entries that have an english string as voice string. A quick count on chinese-simp gives 13 entries that start with a latin character (filtering out the latin alphabeth entries), and some of those have a chinese translation appended (like "Rockbox的版本"), while for chinese-trad I get 283 entries and there don't seem to be any entries that have a chinese translation appended. So this is pretty much the translation file being outdated, and some voice strings being the original english strings.

Maybe you want to update that translation?

(I wouldn't mind someone updating the chinese translation of Rockbox Utility either, since that is currently in a bad state )