Home

Awesome

Deprecated (EOL): This project reached end of life. Back then this SDK was born to address gaps in the official Alexa SDK for Java. Since the official SDK improved a lot and is now in way better shape there is no need to further drive this project.

Join the chat at https://gitter.im/alexa-skills-kit-tellask-java/Lobby Maven central SonarQube Coverage

Alexa Tellask SDK for Java

This SDK is an extension to the Alexa Skills SDK for Java. It provides a framework for handling speechlet requests with multi-variant response utterances organized in YAML files that make it easy to create localized skills. This SDK also lets you build your skill in declarative style and avoids a lot of boilerplate code.

Key features

References

Learn how to use this SDK by having a look into the samples - where you can find reference implementations for multi-language skills. You can also learn from the open-sourced and award-winning Morse Coder skill which relies on Tellask SDK as well.

How to use

Before you start make sure your IDE is set up for annotation processing. (How to for IDEA and Eclipse)

Add below Maven dependency to your project.

<dependencies>
  ...
  <dependency>
    <groupId>io.klerch</groupId>
    <artifactId>alexa-skills-kit-tellask-java</artifactId>
    <version>0.2.3</version>
  </dependency>
  ...
</dependencies>

Prepare a request handler

The request handler does what it says. It reacts on incoming speechlet requests and replies with a response. You could either have your skill implementation in a Lambda function or in a Servlet. Here's how you set up an AlexaRequestStreamHandler for your Lambda function:

@AlexaApplication(applicationIds = "amzn1.echo-sdk-ams.app.c26b1c82...")
public class MySpeechletHandler extends AlexaRequestStreamHandler {
}

You could also override a getter for providing a set of application ids supported by the handler. There's more you can set up for your skill, but let's keep things simple. Here's another example for an AlexaHttpRequestServlet in case you want you skill implementation run on a webserver:

@AlexaApplication(applicationIds = "amzn1.echo-sdk-ams.app.c26b1c82...")
public class MyHttpRequestServlet extends AlexaSpeechletServlet {
}

Prepare your utterance YAML file

Having output speech in code is evil. You will know when you start preparing your skill for another locale. This SDK organizes all the editorial content of your skill in YAML files. You're much more flexible with this approach. You manage those contents similar to how you do it in the Alexa developer console - by using intents and slots. This is a sample YAML file

WelcomeSpeeches: "[Hello|Hi|Welcome|Hey]"

SayWelcome:
  Utterances:
    - "${WelcomeSpeeches} {name} <p>Nice to meet you.</p>"
    - "${WelcomeSpeeches} {name} for using my skill."
  Reprompts:
    - "What would you like to do now?"
    - "How can I help you?"

SaySorry:
  - "Sorry, something went wrong."

SayGoodBye:
  - "[Bye|See you|Good bye]"

SayWelcome, SayGoodBye, SaySorry are response intents you will refer to later on. The first one has two response utterances where the engine will pick one of them randomly. Inside YAML files you can also use templates like WelcomeSpeeches to reuse certain speeches in utterances and reprompts by referring to them with ${templatename}. In templates but also directly in utterances you can have multi-phrases wrapped in square brackets. Only one of them is chosen by the engine randomly as well. That said Alexa got five ways of welcoming a user. The utterances also contain an output slot called {name} which will be resolved by the engine. All aforementioned features are also available for reprompts which you define right below the general utterances.

Create an AlexaLaunchHandler

which handles a launch event whenever your skill is started by the user. You register your handler with an annotation.

@AlexaLaunchListener
public class LaunchHandler implements AlexaLaunchHandler {

    @Override
    public AlexaOutput handleRequest(final AlexaInput alexaInput) throws AlexaRequestHandlerException {
        return AlexaOutput.ask("SayWelcome")
                .putSlot(new AlexaOutputSlot("name", "Joe").formatAs(PHONEME_IPA))
                .withReprompt(true)
                .build();
    }

    @Override
    public AlexaOutput handleError(final AlexaRequestHandlerException e) {
        return AlexaOutput.tell("SaySorry").build();
    }
}

There's only one launch handler per skill. The handleRequest method simply asks you to turn an AlexaInput (providing all necessary information like the session and context object) into an AlexaOutput. The output object is where you refer to the response intent you set up in the YAML file. You also give it a slot with name, value and optionally a slot output format. There's one format for each supported SSML-tag in Alexa so the slot value is not only put into the utterance but is optionally wrapped in an SSML tag.

You don't need to have explicit error handling in handleRequest. Whatever goes wrong with your handler you are given the chance to react on it in the handleError method.

From above example what it sends back to Alexa is something like this:

<speak>Hey <phoneme alphabet="ipa" ph="Joe"></phoneme> <p>Nice to meet you.</p></speak>

Create an AlexaIntentHandler

which handles all incoming intents. Like you did with the launch handler you need to register it with an annotation. Secondly you have to tell it which intent(s) your handler should listen for.

@AlexaIntentListener(builtInIntents = {INTENT_CANCEL, INTENT_STOP}, priority = 100)
public class CancelIntentHandler implements AlexaIntentHandler {
    @Override
    public boolean verify(final AlexaInput input) {
        return true;
    }

    @Override
    public AlexaOutput handleRequest(final AlexaInput input) throws AlexaRequestHandlerException {
        return AlexaOutput.tell("SayGoodBye").build();
    }

    @Override
    public AlexaOutput handleError(final AlexaRequestHandlerException exception) {
        return AlexaOutput.tell("SaySorry").build();
    }
}

You can choose from one to many of the built-in intents of Alexa or you can provide names of your custom intents defined in the intent schema in the Alexa developer console.

Whenever an intent is received from the speechlet handler the intent handler is picked automatically in case its verfiy returns true. This is how you could have multiple handlers for the same intent (e.g. the YES-intent) and let the engine pick the correct one based on certain conditions checked in the verify-method. If there's more than one intent handler interested in the same intent and all of them also verify the request then priority comes into play.

Once again exception handling in handleRequest is done for you from the outside and errors will be routed to handleError so you can react on it with output speech.

AlexaInput

The AlexaInput is given to the launch- and intent handlers. It provides everything which comes with the speechlet request and also gives you an AlexaSessionStateHandler useful for reading/writing state to Alexa session (learn more about state handlers in Alexa States SDK) Most important for you might be the intents coming in with a request. AlexaInput has some really useful helpers so that you can check and get values from input intents.

Assume you work with an AlexaInput in the _verfiy_method of an intent handler, the following might be interesting to you:

@Override
    public boolean verify(final AlexaInput input) {
        final boolean slotIsNumber = input.hasSlotIsNumber("slotName");
        final boolean slotIsNotBlank = input.hasSlotNotBlank("slotName");
        final boolean slotEquals = input.hasSlotIsEqual("slotName", "someValue");
        // find phonetic siblings - best for English language
        final boolean slotEngPhoneticEqual = input.hasSlotIsDoubleMetaphoneEqual("slotName", "drew");
        // find phonetic siblings - best for German language
        final boolean slotGerPhoneticEqual = input.hasSlotIsCologneEqual("slotName", "truhe");
        // find phonetic siblings - picks the best algorithm depending on the current locale
        final boolean slotPhoneticEqual = input.hasSlotIsPhoneticallyEqual("slotName", "truhe");
        final boolean slotHasTrueValue = input.hasSlotIsTrue("slotName");
        // ...
    }

So you can check a slot for a number but also for a certain value. Moreover, you could even check for a phonetic equivalent by levering Double Metaphone or Cologne Phonetic algorithm. For example the line checking for value of "drew" with hasSlotIsDoubleMetaphoneEqual will return true if the slot value contains "true" as a value - which is a phonetic sibling of "drew". However, this is not true in German language but instead the word "truhe" is a phonetic sibling for "true". Finally you can obtain a slots value with getSlotValue.

AlexaOutput

You saw AlexaOutput is returned by the launch- and intent-handlers. There's a lot you can provide to the output:

Provide slot values from POJO models

Instead of explicitly set slot values with AlexaOutput.putSlot you could also give it a model class which extends from AlexaStateModel and got fields with AlexaSlotSave-annotation. Once you give an instance of this model class to AlexaOutput.putState the underlying speechlet will put values into the slots of an utterance. Moreover, the speechlet also takes care of writing state of AlexaStateSave-annotated fields to either Alexa session or one of the persistence stores supported by States SDK. Learn more about POJO state models and AlexaStateHandlers with States SDK

This is a typical POJO state model:

public class Calculation extends AlexaStateModel {
    @AlexaStateSave(Scope = AlexaScope.USER)
    @AlexaSlotSave(slotName = "precision", formatAs = AlexaOutputFormat.NUMBER)
    private int precision = 1;

    @AlexaSlotSave(slotName = "result", formatAs = AlexaOutputFormat.NUMBER)
    @AlexaStateSave
    private double result = 0;
    // ...
}

And this is how you would read/write model state from and to a store. Also the precision value will be filled up in an output slot if it exists in a response utterance:

@AlexaIntentListener(customIntents = "Precision")
public class PrecisionIntentHandler implements AlexaIntentHandler {
    @Override
    public boolean verify(final AlexaInput input) {
        return input.hasSlotIsNumber("decimalplaces");
    }
    @Override
    public AlexaOutput handleRequest(final AlexaInput input) throws AlexaRequestHandlerException {
        final AlexaStateHandler sessionHandler = input.getSessionStateHandler();
        final AlexaStateHandler dynamoHandler = new AWSDynamoStateHandler(sessionHandler.getSession());

        final Calculation calc = sessionHandler.readModel(Calculation.class)
            .orElse(dynamoHandler.readModel(Calculation.class)
            .orElse(dynamoHandler.createModel(Calculation.class)));

        cal.setPrecision(Integer.valueOf(input.getSlotValue("decimalplaces")));

        return AlexaOutput.ask("SayNewPrecision")
            .putState(calc)
            .build();
    }
    // ...
}

This also is a good example of how to subscribe / listen to an intent matching a custom critera (a slot named decimalplaces must have a numeric value).

A corresponding utterance in the YAML file could be the following:

SayNewPrecision:
  - "From now on I round results for you to {precision} decimal places"
  - "Calculation results will be rounded to {precision} decimal places"

Store your utterances YAMLs in an S3 bucket

Another great feature of this SDK is to source editorial content out of the JAR. By storing utterance YAML-files in an S3 bucket you can easily work on speech phrases in your skill without redeploying code.

The SDK uses an UtteranceReader for accessing YAML files. By default ResourceUtteranceReader is set to read YAMLs from the /resources-folder in your project. The default file-structure is the following

/resources
    /{locale}
        /utterances.yml

As an example you might end up with the following

/resources
    /de-DE
        /utterances.yml
    /en-US
        /utterances.yml
    /en-GB
        /utterances.yml

Of course you can alter the file-structure as you want by customizing the UtteranceReader in the speechlet handler. Override the getUtteranceReader method and do something like:

@AlexaApplication(applicationIds = "amzn1.echo-sdk-ams.app.c26b1c82...")
public class MySpeechletHandler extends AlexaRequestStreamHandler {
    @Override
    public UtteranceReader getUtteranceReader() {
        return new ResourceUtteranceReader("/leading/path", "/trailing/path/my-utterances.yml");
    }
}

and you will be able to structure your files like this:

/resources
    /leading
        /path
            /en-US
                /trailing
                    /path
                        /my-utterances.yml
            /en-GB
                /trailing
                    /path
                        /my-utterances.yml

As you now know how to set up an UtteranceReader you can leverage S3UtteranceReader for having your skill load utterances from files you stored in S3. Let me demonstrate how to set this up with the Servlet (you can do the same with AlexaRequestStreamHandler for Lambda)

@AlexaApplication(applicationIds = "amzn1.echo-sdk-ams.app.c26b1c82...")
public class MyHttpRequestServlet extends AlexaSpeechletServlet {
    @Override
        public UtteranceReader getUtteranceReader() {
            return new S3UtteranceReader("bucketName", "/leading/path", "/trailing/path/my-utterances.yml");
        }
}

Above file-structure applies - but now instead of your files being in the /resources-folder they are read from the bucketName bucket. Optionally you can give the S3UtteranceReader an AmazonS3Client to provide custom credentials, another AWS region or some proxy configuration. If you don't, S3UtteranceReader uses the default AWS configuration from the runtime environment.

That's it

Although there's even more to discover in this SDK you already got the most important basics for creating a skill with Tellask SDK. There's a lot going on behind the scenes but you won't care. You are just asked to:

  1. Create a speechlet handler by creating a class either extending AlexaRequestStreamHandler (for Lambda functions) or AlexaSpeechletServlet (for an external webservice)
  2. Built YAML-files with multi-language utterances and reprompts
  3. Create a set of intent handlers which extend from AlexaIntentHandler and register them with just adding the AlexaIntentListener-annotation
  4. Create one launch handler which extends from AlexaLaunchHandler and register it with the AlexaLaunchListener-annotation

Rest of the time you spend on your actual skill logic. Good luck.

More details on this SDK can be found in the Javadocs