Version history

Additions/Changes:

• Support for 64-bit Windows
• Change of license - now free for non-commercial use.
• Several bug fixes
  

TalkingJava SDK Version 1.6.3 (Jan 21st, 2004)

Additions/Changes:

• Added phoneme() method to CGSpeakableListener, which is triggered when a Synthesizer starts to synthesize a new phoneme. There is a corresponding getPhoneme() method in CGSpeakableEvent
• Added valid time stamps (ResultToken.getStartTime() and getEndTime()) for ResultListener.resultRecognized() events.
  

Additions/Changes:

• Added setKeepOpen(boolean keepOpen) method to CGPullBufferDataSource, to force the DataSource to stay open even after zero bytes are written to it. Calling this method with an argument of true allows an RTP stream to function correctly when connected to a Synthesizer which only speaks occasionally - see the examples.rtp.SpeakToRTP class for a demonstration.

• When the com.cloudgarden.audio.AudioMediaURLSink was used to write to a file, the file would keep on growing and never be closed - this was due to the CGPullBufferDataSource's internal "keepOpen" field being set to "true" (see above). Now by default it is set to "false" and for RTP applications it must be set to true using the setKeepOpen method (see above). This bug caused the examples.audio.WAVToMPEG example to never terminate.

Additions/Changes:

• The CGResultAudioClip class (which is the AudioClip returned from the FinalResult.getAudio method) can be saved to a WAV file via the CGResultAudioClip.saveToFile method.
• LicenseManager can now operate via a proxy to validate license, as well as in "off-line" mode.

• In rare cases a SAPI5 RuleGrammar failed to be updated - this has been fixed.
• A few other minor bugs have been fixed.

TalkingJava SDK Version 1.6.0 (Mar 24th, 2003)

Additions/Changes:

• Introduction of name "TalkingJava SDK" to represent the combination of examples, documentation, installer and CloudGarden's JSAPI implementation - the version numbering scheme is continued from older versions.
• Cleaner reorganization of file structure.
• New installer, which allows cleaner management of installation/removal of SDK, and of the JSAPI implementation into JREs and locations where Netscape and IE VMs can access it.
• Examples using the javax.media.rtp package have been added in the examples.rtp package.
• A simple speech-enabled browser example has been added (examples.applet.BrowserApplet) to demonstrate a more sophisticated use of the JSAPI in an Applet.
• A CGPullBufferDataSource can now be created publically from an AudioSource

• Alternative results are now generated for RuleGrammars
• Alternative results are now generated for the Dragon engines for rule and dictation results.
• Several problems preventing certain uses in Applets have been overcome.
• Very occasionally AudioManager.closeOutput() would fail to close the output - now it always does.
• Some problems with using RTP (resulting in occasional data loss) have been solved.
• Long grammar names are now accomodated in SAPI4 engines.

Bug Fixes:

• When loading a JSGF grammar with no "#JSGF..." line or grammar name in the header, a NPE was thrown - this has been fixed.

Release Version 1.5.2 (Oct 27th, 2002)

Additions/Changes:

• The javadoc has been updated for the com.cloudgarden.speech and com.cloudgarden.speech.userinterface packages
• The examples have been more fully documented.
• Much more detail added to reporting of errors in RuleGrammars. (see examples.recognition.GrammarErrorTest)
• GrammarException.getDetails has been implemented to provide more detailed information on all the grammar errors when a RuleGrammar is parsed - instead of just the first error.

• FinalRuleResult.getTags sometimes gave extra tags - this has been fixed.
• Tagging a <NULL> rule is now accepted - eg, this rule: twenty {2} ( <NULL> {0} | one {1} ) will be accepted and converted internally to ( twenty {2} {0} | twenty {2} one {1} )
• Bug causing recognizer to ignore resume method call if a result was recognized while recognizer was paused - this has been fixed.
• Some additional grammar errors caught, eg: <rule> = something * {tag}; //an error is thrown now
• Multiple tags allowed now: eg, <rule> = something {tag1} {tag2} {tag3}; //is OK now
• Escaped characters inside tags are allowed now: eg, <rule> = something {\{tag\}};//is OK now

Release Version 1.5.1 (Oct 20th, 2002)

Additions/Changes:

• If a RuleGrammar with no public rules is loaded into a Recognizer, commitChanges will now throw a GrammarException immediately (instead of a RuntimeException being thrown from the background thread which is commiting the grammar). SAPI4 and SAPI5 engines do not accept such grammars, even though they may be valid JSGF grammars (though such grammars are useless since they have no speakable (public) rules).
• When run in a 1.3 or better JVM, finalizers are not run on exit (instead, a shutdown hook is used for final cleanup). For JVMs 1.2 and earlier, finalizers are run on exit.

• For SAPI5 TTS engines, calling props.setRate(props.getRate()) caused the rate to increase slightly instead of staying exactly the same.
• The SpeechEngineChooser does *not* now call CGEngineCentral.setAWTSynchronization(false) - this lack of synchronization sometimes caused stalling when running SpeechEngineChooser (and may have caused external application code to stall also).
• SpeechEngineChooser now internally creates recognizers which are *not* already running - previously it used already-running recognizers (and deallocated them when it closed) which could have caused problems if a recognizer was used both before and after creating a SpeechEngineChooser.

Release Version 1.5.0 (Oct 6th, 2002)

Additions/Changes:

• Major revision of the com.cloudgarden.audio package.
• AudioMediaFormatConverter allows audio data to be transmitted across a network in compressed format (as opposed to raw data) allowing for much faster interaction with client-side applications.
• Simpler usage of all classes (eg, an AudioLineSource can be created using an AudioFormat, internally creating a DataLine, instead of having to create a DataLine externally and pass it to the AudioLineSource constructor).
• Some renaming of classes -
• AudioPlugSource/Sink becomes AudioClientSource/Sink
• AudioSocketSource/Sink becomes AudioServerSource/Sink
• AudioConverterSource becomes AudioFormatConverter
• AudioSink.startGetting method has been removed - "pumping" of audio data should be carried out with AudioSource.startSending().
• TransferListeners can be attached to AudioSources/Sinks to monitor the passage of audio data
• ClientListeners can be attached to AudioServerSource/Sinks to monitor connection/release of clients and transfer of data.
• AudioFormatConverter can handle conversion between 8 and 16 bit formats - previously limited to 16 bit formats only.
• More examples have been added.
• A few bugs fixed.
• Allowed multiple SAPI4 recognizer instances to be created for the same engine type (this was restricted in version 1.3 in order to support the ViaVoice recognizers, but is no longer needed).

• Some bugs relating to the importing of rules have been fixed.
• The RuleParse.getTags() would often return extra tags - this has been fixed.
• RuleGrammar.parse(String text, String ruleName) did not correctly function when ruleName was null.
• the RuleGrammar.setEnabled(boolean) method now disables/enables all public rules, and if an individual rule is enabled, the RuleGrammar is enabled (as per JSAPI 1.0 specs). Previously if a RuleGrammar was disabled, no rules would be active, even if individually enabled, and when a RuleGrammar was enabled, only those rules individually enabled would be enabled.
• Problems causing native errors or lack of audio data when playing back recorded audio have been fixed.
• The Dragon engine can now accept input from an AudioLineSource.

Release Version 1.4.1 (Aug 19th, 2002)

Additions/Changes:

• Made SpeechEnginePanel navigable by keystrokes
• Added VerbalVision - a screen-reader for Java Swing applications. To use simply install the jsapi files in the lib/ext and bin folders of your JRE as usual, install Sun's Java Accessibility Utilities in the same JRE, and include the following line in the lib/accessibility.properties file: assistive_technologies=com.cloudgarden.speechapps.VerbalVision
• Made "cancelAll" and "cancel" methods quicker to respond when the queue is empty.

• Fixed bug which stopped a grammar from being disabled (bug introduced in 1.4.0)
• Fixed bug which stopped Philips speech engine from commiting a new grammar or changes to an existing grammar after the recognizer started processing speech.

Release Version 1.4.0 (July 19th, 2002)

Additional Engines supported:

• Implementation can now be used with the following additional ASR engines:
• ScanSoft's Dragon NaturallySpeaking (Versions 5 & 6)
• the Philips ASR engines (but note that the Philips engine interferes with IBMs ViaVoice engines, so only one or the other should be installed at the same time).
• Implementation can now be used with the following additional TTS engines
• the L&H TruVoice TTS engines
• ScanSoft's SAPI5 RealSpeak TTS engine.
• Infovox's engines (but due to a bug in the Infovox engines the CGEngineProperties.useFixForInfovox330(true) method must be called to handle wordReached events correctly).

• SynthesizerPropertiesPanel and RecognizerPropertiesPanel added for engine-independent control panels for adjusting engine properties (eg, volume, speaking rate, pitch, sensitivity etc) and for testing voices and profiles.
• SpeechControlPanel redesigned.
• Mouth now implements SpeakableListener so is easier to attach to a synthesizer.
• SynthesizerModeDescriptor now has non-null engineName and modeName fields so that a single SynthesizerModeDesc holds only the voices for a particular Locale, engineName and modeName - previously engineName and modeName were null so that a single SynthesizerModeDesc held all the voices for a particular Locale.
• Java installer changed so that the installation jar file (eg. jsapi10-cg140-free.jar) can be renamed and the installer will still work.

• Various threading issues resolved by an internal redign of message handling.
• Windows 98 bugs fixed (may apply to other OSs too):
• Problem with ViaVoice engines stalling while allocating on Windows 98 has been fixed.
• Various bugs causing TTS engine crashes in Windows 98 have been fixed.
• Bug that prevented grammars from being committed in ViaVoice Recognizers using languages other than English has been fixed.
• Bugs which were causing various problems when getting audio input from files have been fixed.
• Bug that reset a synthesizer's output to the default audio device (if output was previously redirected to a file) when a synthesizer's voice was changed has been fixed.
• Errors corrected in RemoteDictationServer.bat, RemoteDictationClient.bat, RemoteSynthesisServer.bat and RemoteSynthesisClient.bat scripts - they were pointing to the wrong class names.
• SpeechControlPanel bugs corrected.

Release Version 1.3.2 (Mar 9th, 2002)

Additions:

• Implemented the DictationGrammar.setContext methods. Possible uses include allowing the recognizer to make better guesses at what has been said based on words which might not have come from the recognizer, such as written text obtained (from a person, AliceBot etc) in response to text previously obtained from the recognizer.
• Support for <NULL> and <VOID> rules has now been added.
• Added examples.userinterface.VoicePad example - a Swing-based synthesizer application.
• Added TestExamples.bat - a script which starts up a Java app which lets you run 12 of the test demos with output going to a scrolling Java window - useful for those poor soles working on Windows 98 or Me boxes who want to see more than 50 lines of text.

• Lengthened demo trial period from 30 days to 60 days.
• JSGF Grammar-definition errors are now reported more helpfully.
• Changed name of DLL to cgjsapi132.dll (ie. it is now version-dependent) which will remove problems caused by loading the wrong version of the dll.
• The DictationGrammar methods addWord, removeWord, listAddedWords and listRemovedWords throw RuntimeExceptions with a message to use the almost-equivalent VocabManager methods instead, since the VocabManagers methods are generally more consistent with this SAPI4/5-based implementation.

• Fixed bug which caused Synthesizer to hang sometimes if asked to speak something which made no sound - eg: synth.speak("",null) could sometimes hang the Synthesizer.
• Fixed bug which prevented a new DataSource object from being obtained from the CGAudioManager even after the CGAudioManager.closeOutput() method was called. The first one created was always returned. A similar bug applied to getting a new DataSink object.

Release Version 1.3.1 (Jan 14th 2002)

Additions:

• The VocabManager getWords method: for SAPI4 SR and TTS engines (presently only the Microsoft engines support this feature) now returns the IPA pronunciation of an arbitrary (valid) word, and not just one added by using the addWord method - see the examples.vocab.VocabTest example for a demo.

• SpeechEngineTree: - if the "Recognizers" node was double-clicked all the recognizers were replaced by the two profiles of the first recognizer, also causing a ClassCastException to be thrown when the SpeechEngineTree was closed. The Recognizer node now expands/collapses in the normal way.
• RuleGrammars: if more than one grammar was imported at the start of a grammar file, all but one of the grammars would not be imported correctly, and the rules of those grammars would not be found. Imports now work correctly.
• Various smaller bugs which generally would have reduced recognition performance and parsing of recognized and arbitrary text. Basically, if you were having problems with recognition (especially with SAPI4 engines) please upgrade to this version.

 

Additions:

• Support for SAPI4 Recognizers - (including IBM's ViaVoice, and Phillips engines). Now all SAPI4 and SAPI5-compliant recognition and synthesis engines are supported. Note that though some engines claim to be SAPI4 or SAPI5-compliant they do not necessarily support all features of SAPI4/5 - and so, for example, not all JSML tags will be interpreted correctly by certain TTS engines, such as the ViaVoice or L&H engines. On the whole, though, compliance is very good.
• Lip-sync events Lip-sync events are now detected from Synthesizers and CGSpeakableEvents broadcast to CGSpeakableListeners, which can then display the current shape of a mouth using the com.cloudgarden.speech.userinterface.Mouth Component.
• Confidence levels returned with FinalResultEvents Recognition results are given with values representing the confidence with which they have been recognized by the speech engine, allowing a certain degree of feedback to provided in language-training applications.
• ResultEvent.audioLevel method implemented (standard JSAPI 1.0, but just not implemented previously)
• com.cloudgarden.speech.userinterface package - including a "Mouth" Component for displaying lip-sync events, and a SpeechEngineChooser dialog for testing and selecting speech engines, profiles and voices.

• Engine names changes - addition of "SAPI4" or "SAPI5" and manufacturer to basic name, eg "Mary, SAPI4, Microsoft"
• Removal of limit on String-length of synthesized speech - in previous versions there was an (undocumented) limit of about 2500 characters on the length of a String sent to a speech synthesizer. This has now been removed and Strings of any length are handled correctly.
• When writing to files, the Synthesizer "speak" methods now behave asynchronously and the usual SpeakableEvents are sent, (as opposed to previous versions, which wrote synchronously, returning only after all data had been sent, and produced no events).
• Various small bugs relating to committing changes to grammars and getting alternate text for non-current results have been fixed.

Changes:

• Integration with the JMF - getDataSource and getDataSink methods have been added to the CGAudioManager class, allowing full integration with Sun's Java Media Framework. This allows, for instance, speech data to be transmitted or saved in MPEG or other compressed audio formats, and for audio data to be transmitted across a network using the JMF (though this can also be done using other classes in the com.cloudgarden.audio package).
• Field values changed - Values of public static fields (such as Engine.DEALLOCATED) have been changed to equal those of IBM's Java Speech for ViaVoice, so that code compiled using IBM's JSAPI will work with Cloudgarden's JSAPI without requiring recompilation. However, this means that any code previously compiled with Cloudgarden's JSAPI versions prior to 1.2.1 will need to be recompiled to work with version 1.2.1
• Deadlock problems alleviated - Speech engine events for individual speech engines can now be started each in a new Thread (this behaviour can be controlled by using the CGEngineProperties.setEventsInNewThread method) so that waitEngineState can now be called from an event handler method without causing deadlock. See the examples.DeadlockTest code for a demonstration of deadlock and how to avoid it.
• VocabManager.listProblemWords is now implemented, and bugs involving the SAYAS tag and SAPI4 TTS engines have been fixed.
• The implementation is now provided in files called "cgjsapi.jar" and "cgjsapi.dll" (in previous versions the files were "jsapi.jar" and "jsapi.dll", which could have led to ambiguity problems).

 

• The length of audio files is now specified in their headers so that they should be able to be read by any audio file player.
• VocabManager.listProblemWords now implemented
• Problems encountered when disabling and re-enabling entire, or parts of, RuleGrammars have been fixed.

Release Version 1.2 (October 14, 2001)

Additions:

• Support for SAPI 4 as well as SAPI 5 TTS engines. All SAPI 4 and SAPI 5 TTS engines installed on a machine are detected and any of them may be selected and used for speech synthesis.
• The com.cloudgarden.speech.audio package, which offers a flexible scheme of AudioSources and AudioSinks (with AudioFileSource, AudioLineSource etc implementations) to allow Synthesizer output to be redirected to Files, javax.sound.sampled.Lines, remote client machines or custom AudioSinks, and similarly, Recognizer input to come from Files, audio Lines, remote clients or custom AudioSources.
• Is now Java 1.1 compliant, so can be run from any Applet in a suitable browser (eg Internet Explorer 5.5), after suitable installation of both the JSAPI and the Java Sound package, or Java Media Framework. Note that this Applet usage does not require installation of a separate JRE since it uses the browser's VM. It can also, of course, be run using the Java Plugin, if the JRE 1.2 or 1.3 is installed.
• The JSAPI is less susceptible to deficiencies in almost-SAPI4-compliant engines.
• There are several examples to demonstrate using remote clients for recognition and synthesis, as well as IO to/from audio files and Lines.

• The (non-JSAPI-spec) methods previously added to the AudioManager have been removed and are available instead under the com.cloudgarden.speech.audio.CGAudioManager class, along with the other methods and classes mentioned under "Additions"
• Likewise, other non-jsapi methods previously added to the AudioManager, SpeakerManager and EngineProperties classes have been moved to the CGAudioManager and CGEngineProperties classes, leaving the implementation true to the JSAPI 1.0 specs, apart from the addition of the Word constructor, and the recognizerListening method in the RecognizerAdapter class.
• Synchronization with the AWT EventQueue is now controlled by a call to the setAWTSynchronization(boolean) method of the com.cloudgarden.speech.CGEngineCentral class. Thus it can be turned off in Applets allowing them to run in Netscape and Internet Explorer, and also without needing to give them permission to write to a System Property.

• The Synthesizer.cancelAll() method has been fixed.
• Non-English speech engines are correctly detected and assigned to SynthesizerModeDescriptors.

Additions:

• Some new examples were added to test for duplex card capabilities (computer speaking and listening at the same time) and to list speech engines.

• The speech impediment has been removed for the trial version and been replaced by a 30-day evaluation period. The evaluation version is now fully-functional and so it is now also possible to add words with pronunciations to the VocabManager.
• setOutputFile and closeOutputFile now throw an IOException if they are unable to access the file (a possible reason being that the file is already open by another application such as a web server or a media player).

• The conditions resulting in native code exceptions, thrown previously when an engine shut down incorrectly, or when there was only one speaker profile, have been fixed.
• The FinalRuleResult methods getAlternativeTokens(int nBest), getRuleGrammar(int nBest) and  getRuleName(int nBest) now return the best-guess values for nBest = 0 (and null for values > 0, since the Microsoft engines do not support alternatives for RuleGrammars).
• A new SynthesizerModeDesc is created for every group of voices for a given Locale, so that now the Central.availableSynthesizers method returns an English SynthesizerModeDesc with four Voices (Mary, Mike, Sam and the Sample voice), a Chinese SynthesizerModeDesc with the Simplified Chinese voice (which unfortunately just spells out words), and a Japanese SynthesizerModeDesc with the Japanese voice. Previously all voices were added to a single SynthesizerModeDesc.

Additions:

• Implemented SpeakerManager methods:
• getCurrentSpeaker()
• setCurrentSpeaker(SpeakerProfile)
• listKnownSpeakers()
• getControlComponent()
• getControlComponent(int type) - non-JSAPI-spec method allows several types of control component to be accessed.
• Implemented EngineProperties method:
• getControlComponent()
• getControlComponent(int type) - non-JSAPI-spec method allows several types of control component to be accessed.
• Added VocabManager methods
• addWord(Word) and addWords(Word[]) - note - pronunciation is used only in the commercial version.
• removeWord(Word) and removeWords(Word[])
• Added Word constructor
• Word(String writtenForm, String spokenForm, String[] pronunciations, long categories) - non-JSAPI-spec method so that new words or new pronunciations can be added to the VocabManager
• Added FinalResult methods:
• getAudio() - returns an AudioClip whose only working method is play() i.e. stop() and loop() are not implemented
• getAudio(int, int ) - non-JSAPI-spec method, gets alternative tokens, identified by their position rather than their value, since the same token may appear more than once in a FinalResult
• isAudioAvailable()
• isTrainingInfoAvailable()
• releaseAudio()
• releaseTrainingInfo()
• String[] getAlternativeTokens(int fromPos, int toPos, int maxAlts) - non-JSAPI-specmethod, gets alternative tokens, identified by their position rather than their value, since the same token may appear more than once in a FinalResult
• tokenCorrection(String[] correctTokens, int fromPos, int toPos, int correctionType) - non-JSAPI-spec method, corrects tokens,  identified by their position rather than their value, which must be contained in the set of tokens returned by the getAlternatives method above.
• Added examples.userinterface.TestInterfaces to demonstrate the native user interfaces available by calling getControlComponent(int type)
• Modified DictationTest, and TestResultListener to demonstrate getAudio, set/getCurrentSpeaker, and tokenCorrection methods

• Creating multiple Synthesizers resulted in either incorrect handling of events or crashes in the native code. This has now been corrected.
• The limits of 10 Synthesizers, Recognizers, and also 10 Grammars per Recognizer, have all been lifted.
• Tags may now have spaces without causing a grammar exception

Release Version 1.0 (May 20, 2001)

Additions:

• AudioManager.setOutputFile and AudioManager.closeOutputFile, for saving audio output to wave files.
• Added JSAPI Installer to install jsapi files and run demos.

• RuleGrammar.parse(String text, String ruleName) - now correctly fails if the text string does not exactly match against a rule - before the text could match a rule but have unmatched text on the end and still return a non-null Rule.
• Also, a null ruleName now produces the correct behaviour.

• If an EngineModeDesc with running == Boolean.FALSE is specified in createEngine a new engine is now always returned.
• When multiple Recognizers were created they did not process results in parallel, but they do now.

• RuleGrammar.ruleForJSGF - incorrectly looked for rule declarations like public <rulename>= "definition" and not just the rule definition, so would throw an exception when given a rule definition like "hello [world]"
• RuleGrammar.listRuleNames - fixed UnsupportedOperationException
• RuleParse - fixed case where there are multiple branches of a RuleAlternative all starting with similar phrases - bug would return a summation of result tokens, and not just the single correct alternatives