[mary-users] Distortion and buzzing

[Marc Schroeder]

Chuck,

ah, the marvels of working with computers... just when you thought the 
basics are working in a trustworthy way, a software upgrade proves you 
wrong... :-)

What you describe sounds to me like a bug in the Java sound 
implementation in the Java 5 version shipping with Leopard. I will be 
upgrading my Intel MacBook Pro to Leopard some time early next year, so 
I will be able to witness the problem myself...

... but if you want to pinpoint the problem separately from the MARY 
system, you could use the command-line audio player that comes with 
mary's signalproc.jar:

java -cp ".../MARY TTS/java/signalproc.jar" 
de.dfki.lt.signalproc.util.AudioPlayer myfile.wav

If that shows the same problem, then we know it's got nothing to do with 
TTS as such. Next thing I would try is converting the mono file to 
stereo with some external software (e.g., Audacity), and trying to play 
back that one. Next thing, use a sample rate natively supported by the 
audio hardware (44100 or 48000 Hz)... enough to keep oneself busy.

If you feel really exporatory, you could try with the Java 6 developer 
preview release, and see if the problem is still in that one...


Anyway -- for now, best wishes for a Merry Christmas and a Happy New Year!

Marc



Charles Rich schrieb:
> 
> ------------------------------------------------------------------------
> 
> 
> Hi Marc,  Thank you for your prompt and thorough response.   I have done 
> some further research based on your suggestions
> and discovered the following interesting results:
> 
> * Same problem for all three types of voices.
> 
> * When I save a WAV file from the client, it sounds fine played back 
> (with QuickTime).   That would suggest the problem is in the client.
> 
> * The voice is still intelligible.  Please listen to attached recording 
> of hmm-slt saying "Welcome to the world of speech synthesis!", which I 
> made by holding a microphone to the *right* speaker on my Mac (see 
> below) and using Sound Recorder on my PC.
> 
> But here's the real surprise:
> 
> * I noticed that the sound from the *left* speaker is ok, but the sound 
> from the *right* speaker is bad.   It is *not* my speaker or sound board 
> hardware, because when I play music (and shift the balance back and 
> forth to check), everything is crystal clear.
> 
> * Before I resorted to holding a mic next to the speakers, I tried using 
> Audio Hijack Pro (http://www.rogueamoeba.com/audiohijackpro) which uses 
> SoundFlower (http://www.cycling74.com/products/soundflower) to directly 
> record the system audio output from the Mary client.   However, when I 
> did this, the audio was good! 
> 
> Some further  thoughts:
> 
> (1) Does someone around your lab happen to have a new Intel 
> dual-processor MacBook Pro (running Leopard and JSE 5.0) on which you 
> could try to reproduce the behavior I am seeing?  (Or just Leopard on 
> any machine?)
> 
> (2) I have exactly the same symptoms with FreeTTS.   Does this suggest 
> anything?
> 
> (3) I assume that the basic speech synthesis is mono---how is the right 
> channel generated?
> 
> (4) I encountered a problem with a game engine in Leopard related to the 
> fact (I was told) that Leopard finally switched its audio support from 
> the old Cocoa system to something called coreaudio.    Does this mean 
> anything to you?  How does the final binding of Java to the OS X audio 
>  happen?
> 
> (5) Would it be appropriate to try the "native" libraries?
> 
> Thanks, -Chuck
> 
> On Dec 20, 2007, at 4:13 AM, Marc Schroeder wrote:
> 
>> Hi,
>>
>> is this the case for all voices, or just certain ones? Have you tried:
>>
>> a) unit selection voices (e.g., slt-arctic);
>> b) hmm-based voices (e.g., hmm-slt);
>> c) mbrola voices (e.g., mbrola-us1)?
>>
>> Is the buzzy sound only when playing back in the GUI client when 
>> pressing the "Play" button, or also when you save as WAV? (i.e., is 
>> the problem in the server or the client)
>>
>> Is the speech still intelligible, or not? Can you send an example?
>>
>> It sounds to me like a byte-order problem, but without more detailed 
>> information, it is difficult to know.
>>
>> Regards,
>> Marc
>>
>> PS: MARY TTS/lib/native is present only when you install native 
>> components, e.g. the english-shprot component in the installer. This 
>> should be unrelated.
>>
>> Charles Rich schrieb:
>>> Hi, I have just installed the latest version of Mary on my new 
>>> MacBook  Pro (running Leopard 10.5.1 and Java 5.0).   When I run the 
>>> browser  demo at http://mary.dfki.de/online-demos/speech_synthesis, 
>>> it sounds  fine, but when I run my locally installed server and 
>>> client, I get  terrible distortion and buzzing.   I thought at first 
>>> something was  wrong with my speakers or sound card, but that is 
>>> obviously not the  case, since the online demo is pure and clear.
>>> Any suggestions?
>>> Thanks, -CR
>>> P.S. I am evaluating Mary for use in student projects for a grad 
>>> class  on intelligent user interfaces I am teaching this spring 
>>> (paired with  Sphinx4 for reco).
>>> ---------------------------------------------------
>>> Dr. Charles Rich, Professor of Computer Science
>>> Interactive Media and Game Development
>>> Worcester Polytechnic Institute, Fuller Laboratories 143
>>> 100 Institute Road, Worcester, MA 01609-2280
>>> Email: rich at wpi.edu <mailto:rich at wpi.edu>   Phone: 508-831-5945 
>>>   Fax: 508-831-5776
>>> Home: http://www.cs.wpi.edu/~rich
>>> _______________________________________________
>>> Mary-users mailing list
>>> Mary-users at dfki.de <mailto:Mary-users at dfki.de>
>>> http://www.dfki.de/mailman/listinfo/mary-users
>>
>> -- 
>> Dr. Marc Schröder, Senior Researcher
>> DFKI GmbH, Campus D3_2, Stuhlsatzenhausweg 3, D-66123 Saarbrücken, Germany
>> http://www.dfki.de/~schroed
>> --
>> official DFKI coordinates:
>> Deutsches Forschungszentrum fuer Kuenstliche Intelligenz GmbH
>> Trippstadter Strasse 122, D-67663 Kaiserslautern, Germany
>> Geschaeftsfuehrung:
>> Prof. Dr. Dr. h.c. mult. Wolfgang Wahlster (Vorsitzender)
>> Dr. Walter Olthoff
>> Vorsitzender des Aufsichtsrats: Prof. Dr. h.c. Hans A. Aukes
>> Amtsgericht Kaiserslautern, HRB 2313
>>
> 
> ---------------------------------------------------
> Dr. Charles Rich, Professor of Computer Science
> Interactive Media and Game Development
> Worcester Polytechnic Institute, Fuller Laboratories 143
> 100 Institute Road, Worcester, MA 01609-2280
> 
> Email: rich at wpi.edu <mailto:rich at wpi.edu>   Phone: 508-831-5945   Fax: 
> 508-831-5776
> Home: http://www.cs.wpi.edu/~rich 
> 
> 
> 
> 
> 

-- 
Dr. Marc Schröder, Senior Researcher
DFKI GmbH, Campus D3_2, Stuhlsatzenhausweg 3, D-66123 Saarbrücken, Germany
http://www.dfki.de/~schroed
--
official DFKI coordinates:
Deutsches Forschungszentrum fuer Kuenstliche Intelligenz GmbH
Trippstadter Strasse 122, D-67663 Kaiserslautern, Germany
Geschaeftsfuehrung:
Prof. Dr. Dr. h.c. mult. Wolfgang Wahlster (Vorsitzender)
Dr. Walter Olthoff
Vorsitzender des Aufsichtsrats: Prof. Dr. h.c. Hans A. Aukes
Amtsgericht Kaiserslautern, HRB 2313