shithub: flite

ref: b273a40041a0330bd1545316eed63959a6cc4bdd
dir: /wince/flowm.notes/

View raw version

flowm
   Flite on Windows Mobile
   flite file reader development  "flown" or "flowm"
   build up with cegcc menu example
   type in text first
   file selection.
     good searching, bookmarking, replay (slower)
   on screen word highlighting (default -- but optional)
   wget data 
     interpretation of xml marked up programmes/channels
     to get data and read it (like rss)
   skinable 
   scriptable (through ssml)

  Would be good to have voice choices
     CG and VT if possible
       computationally too expensive

0.7
   TODO
*     screen off option on main screen
*       (makes it battery efficient)
*       http://msdn.microsoft.com/en-us/library/ms838354.aspx
*       http://forums.microsoft.com/china/ShowPost.aspx?PostID=4273639&SiteID=15
*     no ending on ok
*    fast backskip causes a crash
*    long filenames badly displayed
*      added ES_AUTOHSCROLL to field
*    more control in play dialog box
*      choose voice
*      display pronunciation/text expansion etc
*    allow waveform saving in play
*    BM in play text not play file
*    normalize volume across voices (boost cg ones)

     do something with silences to optimize speed
     back/forward over restarts seems to over skip
     keyboard accelerators (at least ctrl-q)
     can't stop BM "1982 was ..." would take a little time
     get rid of title bar on open/savefile box
        tgps doesn't have the problem
     "save" grey out if nothing to save
     some content on the startup screen
        logo plus play or play_file buttons
     CAB installation (with lcab)
        no evidence lcab works for mw yet
        http://nedge2k.com/cab-o-matic ??
     cut and paste from other apps
     file position over app restarts
         and voice selection
     make it not look like a .net app
     screen off after n seconds of play
     previous files selection in file selection
     button support (no ending on ok, stop continue etc)
     screen rotate
     icons rather than text on buttons
         play/stop as one button
         autorepeat on forward/back
         scroll wheel (volume or forward back or faster/slower)
     dll version (for voices too) would be good
     progress bar
     word highlighting at play time
     a talking head example
     prc files
     Inertia based file listing 
       Inertia based position finding in file

0.6 19th
   5.01meg with kal, awb, rms and slt

   with screen off, 40 minutes of running used 6% of battery
      (kal might be different from cg voices)
   a benchmark in flowm so I can find out the speed of the machine
      GetSystemTime (but no milliseconds)

   if #define double float in cst_mlsa.c
      Tilt kal 503.5  16.0 31.47
           awb  44.3  50.0  0.89  but it *is* fast enough (sort of)
           rms  48.9  54.0  0.91  
      and a wee bit faster again if wifi is turned off 

      Tilt kal 503.5  15.0 33.56 
           awb  43.7  69.0  0.63
           rms  48.3  72.0  0.68
      2755 kal 503.5   8.0 62.93 
           awb  43.7  52.0  0.84 1.50
           rms  48.3  53.0  0.91 1.50
      3635 kal 503.5  23.0 21.89
           awb  43.7 118.0  0.37
           rms  48.3 123.0  0.39
      P2   kal 503.5  12.0 41.96
           awb  43.7  68.0  0.64
           kal  48.3  70.0  0.68
      
   dll version (for voices too) would be good
   some content on the startup screen
   - forward/back utterance
         forward/back paragraph/chapter/file
     quick back skips causes crash
   - try out PowerPolicyNotify(PPN_UNATTENDEDMODE, 1); during play
        http://channel9.msdn.com/wiki/mobiledeveloper/powerpolicynotify/
   * do file pos as percentage
         progress bar
   get rid of title bar on openfile box
   * cg voices give click at buffer change
     good enough on 2755 (rms and awb order 13/8K)
     * get a 641K awb voice (and others)
       720K actually with 13 params
   * good enough for tilt (#define double float)
   * multi-voice version
   * made audio_streaming_info with min_buffsize
   file position over app restarts
   icons rather than text on buttons
       play/stop as one button
   make it not look like a crude app
   cut and paste from other apps
   more control in play menu
      display pronunciation/text expansion etc
   word highlighting at play time
   a talking head example
   keyboard accelators
   ok should background it not stop it

0.5
   - forward/back utterance
         forward/back paragraph/chapter/file
     quick back skips causes crash
   - try out PowerPolicyNotify(PPN_UNATTENDEDMODE, 1); during play
        http://channel9.msdn.com/wiki/mobiledeveloper/powerpolicynotify/
   * do file pos as percentage
   cg voices give click at buffer change
     good enough on 2755 (rms and awb order 13/8K)
     * get a 641K awb voice (and others)
       720K actually with 13 params
   * multi-voice version
   * made audio_streaming_info with min_buffsize
   13th January

   hp2 9:54 awb 8k 2755 (black clip) 98%
     backlight off after 2 minutes
   10:30 fp  29500 79%
   12:02 fp 104204 36%  (looks like nearly 3hrs on one charge)
   12:43 fp 136282 16% 

0.4
   using threads so that stop works
     can't share alloced memory over threads
   add streaming synthesis
     stop works
   need to select position in file
   nees to save position in file for continuation
   pause and continue
   stop it from going to sleep (but do screen blank)

   8K awb 1139896
   16K awb 2227760
   could be 641K if order 18 and get rid of std in statics

   Put back in the Play_thread.  With a large buffer size, it sort of
     works (spool vs streaming)

   Tested under 
     Ipaq 3635 WinCE 3.0
     Phraselator 2 Windows CE 4.20 (HPC?)
        gives an error at exit time (maybe the others do too ..)
     HP hx2755 PocketPC 2003 
     AT&T Tilt PocketPC 6.0

  4th Jan 2009

0.3
  TTS from a .txt file
    display tokens in window as we go
  redo the menus and forms layout
  Still very basic though.
  stop doesn't work bacause things don't get back to the event loop.
  2nd Jan 2009   

0.2
  have windows, a dialog box with a play and editable text box
  spent ages working out how to program in WinCE but seems to work now
  1st Jan 2009

0.1
  play_wave(w) does the right thing
     au_wince.c seems to work
  moved to mingw32ce version 
  added a .rc resource file and tidied up configure.in

  moved to cst_file_wince.c rather than cst_file_cegcc.c (deleted)
  fixed bugs in cst_file_wince.c

  type in box and audio out
 
  want flite.dll cmu_us_kal_diphone.dll usenglish.dll maybe   
  have them loaded (so voice isn't in flowm)

  Need examples/tutorial on building an app with inputs, defaults, options
  file loading display etc.

  streaming synthesis? (and call back)

0.0
  compiled flite with message box displaying number of bytes in
  synthesized waveform.  Synthesis seems to work, WCHARs support is
  necessary
  cst_wchar.[ch]
  6th June 2008

-------------------------

Making dlls (using dlls)

  http://craigshaw.com/ (for ffmpeg)

Example code
  http://www.codeproject.com/KB/mobile/

dll building
  http://www.ibm.com/developerworks/aix/library/au-porting/index.html

Books

Programming Microsoft Windows CE, Second Edition (Pro-Developer) (Paperback)
Online sample chapters
http://www.microsoft.com/mspress/books/sampchap/5461b.aspx

code samples (for eVC++)
http://msdn.microsoft.com/en-us/library/bb158662.aspx

Simple, but larger CE example in C++
http://www.codeproject.com/KB/mobile/ce_cards.aspx

join codeproject
awbcmu
publicpass

But found good examples that are (mostly) compilable in 
  Programming Microsoft Windows CE (Doug Boling)
Bought second hand on Amazon for $11
It has online examples, and a version of the book all in CHM format

Its actually all online 
http://iptvforum.cn/wince/wince-programming-ms-press2/source/prowice/ch04d.htm


Keeping app running while in sleep model

            PowerPolicyNotify(PPN_UNATTENDEDMODE, 1);
            SystemIdleTimerReset();        // call this periodically
            SetSystemPowerState(null, POWER_STATE_ON, POWER_FORCE); 

http://forum.xda-developers.com/showthread.php?t=379366