Vocoding fun


I posted a question on StackOverflow a while ago trying to learn the easiest way to get MapQuest to give you turn by turn voice output from Optimus Prime. Yesterday I saw a recent comment inquiring whether or not I had a solution. Today I have a possible solution. I found an open source vocoder written in C (Zerius Vocoder) that I’ve managed to build/run on the Mac. It works pretty good, I must say. Hi, I’m Cliff. You’re here because you obsess over little robot toys that fold into/out of vehicle form. I’m here to report status on what could be a nifty little iPhone app.

Imagine for a moment that you could plug a vocoder into the pipeline of CoreAudio’s queue services API. (Imagine that there were such a pipeline!) That would allow all sorts of interesting voice output! So far I only have a compiled binary with sample files and a rough understanding of what vocoding actually means. Obvious next steps are to finish my Speex tutorial series and refine the hideousness of iPhone audio APIs. I’m really thinking of using pipelines similar to how GStreamer works. If I get distracted reading that link to GStreamer I may end up on a tangent, trying to port GStreamer to CocoaTouch. That’s all for now, party people. I’ll hit you back when I actually have something tangible to share.

OSX Tcl/Tk and snack


So I finally figured out how to install snack on OSX. I went down many different paths and ended up building from source and getting lucky. I’m sorry, what is snack? Apparently it’s some audio tool kit for scripting engines like Tcl, Python, and Ruby. Woah, I’ve probably lost you. Let me back up. Why am I messin’ with Tcl/Tk and snack? I’ve been trying off and on to build a custom voice for MaryTTS. It’s sortr of a pain b/c I don’t have time to mess with this in the office. I’d like to experiment at home but my development linux install is in the office and I haven’t figured out how to get audio over remote desktop on Linux yet. (I’ve gotten close but that’d take a whole ‘nother round of experimenting!) So I’m running OSX on my Mac of course for my MaryTTS tinkering but it’s difficult to get things intended for Linux to work the same way on Mac. I’ve installed Mac Ports which dumped a bunch of stuff somewhere on my hard drive. I managed to get sox from it. Wait, let me start over again! Hi, I’m Cliff. You’re here because you were having the same troubles installing snack on your Mac. I’m here because the prior sentence creates an interesting rhyme.

In experimentation with Mary custom voices I’ve learned that you can spend an entire month just getting the required dependencies for using the voice import stuff. The list of required tools is huge and I wish it could be simplified. To add on that, each tool spins its own dependency web. For eg, installing ehmm (or was it speech tools?) requires sox which requires Mac heads to go out and install Mac Ports or something similar. Running “./configure” for hts just to create the dumb make file requires tcl/and snack. God help you if you’re on modern 64bit hardware like I am. The web site for snack offers a PPC version of snack. It’s best (and painful) to build from source.

So I download the source tar and carefully follow the steps in the README which are incorrect. According to the README the tcl framework should be located in “/Library/Frameworks/Tcl.framework/Headers”. On my system I found it under “/System/Library/Frameworks/Tcl.framework/Headers”, though this might vary depending on how you install Tcl. And that’s the problem. With all the different paths for frameworks and libraries it’s a wonder you can get anything to work! After getting no satisfaction after running “make install” from the source bundle and many failed attempts to copy the libraries into the different Tcl and Tk framework folders I stumbled across a post somewhere that hinted that Tcl extensions should be found/loaded in “/Library/Tcl”. You would never know this unless you were a Tcl veteran. There wasn’t even a “/Library/Tcl” folder on my system! I created the structure and voila! Where was I? Oh yeah, I was trying to install the hts package. Now I’m struggling with a another dumb error that complains it “can’t find swab of SPTK”. What in blue blazes is SPTK and why do I need a swab of it? In conclusion, to build your own voices in Mary you have to be Tcl’ish, have a snack, get a swab of SPTK, oh and make sure you’re wearing sox! That’s only scratching the surface.

Compile Speex For iPhone


Speex is an audio codec specially designed for voice audio. I’m sorry, a codec is a software component used for encoding and decoding something. Oops, I’m terribly sorry! An iPhone is kinda like a phone but it has no wires, y’see… and the letter i is prefixed to it because… Hi, I’m Cliff. You’re probably here because you already know what an iPhone is. Maybe you’ve seen one of the commercials. Maybe your cousin’s best friend’s uncle’s nephew’s best friend’s cousin has one. (In case you missed it, that was a really really verbose way of saying “your cousin”!) Maybe you’ve been around the block once or twice and happen to know about both codecs and Speex. (Maybe it was dem fools throwin’ bones by the liquor store that taught you about Speex. Don’t laugh, cause you’d be surprised by what you learn on the streets these days.) Whatever the case, I’m going to explain how I got a clean compile on the Speex library using XCode. Hold tight because the remainder of this post is designed to be informative… that is all jokes and nonsense to the side.

  1. Download the Speex C source bundle
  2. Unpack it
  3. you might wanna try building it from the cmd line. Run “./configure;make” from the folder where you unpacked it.
  4. Create a new project in XCode. This can be an iPhone or an OSX project but for consistency’s sake (and because it’s what I put in the title) let’s use an iPhone project.
  5. Create an actual folder, not an XCode group, under the root of your new project and call it “CSource”. This is where we’ll put the Speex source code.
  6. Drag/drop the folder into your XCode project or create a group that points to that folder.
  7. Double click your project icon in the left hand tree in XCode to edit the project settings.
  8. Goto the build tab and type “header” in the search field to filter your choices to the things that deal with headers. (Yes I just blatantly included a “goto” in a modern day technical writeup.)
  9. Look for the “Header Search paths” build setting. It should fall under the search paths category somewhere toward the middle of the screen. If you don’t see it finish typing the term “header search paths” in the search bar above. It does an incremental search as you type.
  10. Double click the Header Search Paths build setting to bring up the edit dialog then double click in the value field and set the value to “$(SRCROOT)/CSource”
  11. Open the folder that you unpacked Speex to in Finder and drag the “libspeex” folder directly into the CSource grouping you created in XCode. Choose yes to copy the files.
  12. Back in Finder, navigate to the include folder under the Speex unpack directory. Drag the “speex” folder out of here and next to the libspeex folder you added to your project in XCode. Choose yes to copy the files.
  13. Back in Finder, drag the “config.h” file out of the root of the speex unpacked folder and into your XCode project.
  14. Back in XCode, hit Cmd+Shift+D and type “arch.h” to find and open the arch.h file we’ve imported into our project.
  15. Add a “#include” to include “config.h” at the top.
  16. Remove the “echo_diagnostic.m” file under libspeex from the project as it will just cause complications.
  17. Hit compile and wait for the errors to roll in!

If you followed the above steps correctly then you should only see a few errors relating to duplicate symbols. If you get thousands of errors then it’s likely related to missing header files. You probably have to make sure you imported the speex folder with all the headers and double check your header search path to make sure that it points to the directory containing the imported “speex” folder. You might find a bunch of errors if you don’t remove the “echo_diagnostic.m” file from compilation. The last order of busniess would be to import the ogg container source.I downloaded libogg-1.1.4 which appears to work with speex-1.2rc1. Including the “ogg.h”, “os_types.h”, “bitwise.c”, and “framing.c” files allows me to compile code included from the “speexen.c” and “speexdec.c” examples.