When you carry on a conversation with another person, you can change the subject at any time and continue on smoothly. Our aim is to make that true when speaking with Alexa as well. Alexa should always be responsive to customer interaction.
An Alexa-built in device prioritizes customer interaction over everything else, ensuring it's always able to detect a customer's utterance.
When interactions with Alexa overlap or conflict, in order to make the exchange as smooth as possible, we must think about the order of priority of the interactions. What do we do when multiple conflicting events occur, either from customer request or from the Alexa service? Providing a consistent, intuitive interaction model will increase a customer's engagement with your device, and lead to a much more satisfying overall experience.
Alexa interactions can be divided into these basic groups listed in order of precedence:
Customer-initiated input (for example, utterance or Action button)
Alexa TTS responses (for example, responses to queries or skills)
Alerts
Notifications
Audio/Music.
In order to demonstrate how these situations should be handled, we include the examples below. They will help you understand how to handle a number of interruption scenarios.
For more technical information about how audio channels interact, see Interaction Model.
Important
In whatever way your device handles interruptions, the customer must always be able to dismiss activities, interactions, and responses.
Pausing audio playback
When a higher priority activity interrupts audio playback on your device, you should pause the current activity. For example, imagine a customer is listening to music and asks Alexa what time it is. The media should pause until Alexa's response is complete.
You may choose whether your device resumes or restarts interrupted audio content after an interruption.
Example scenarios
The scenarios in this section are provided to give you specific examples of conflicting Alexa activities and how they should be prioritized. In these cases, and in other use cases that may be unique to your device, we recommended prioritizing interruptions as listed in the Activity types section above, always giving preference to customer-initiated interactions.
Customer interrupting Alexa
When Alexa is speaking
Alexa is reading the 7-day forecast, or any other TTS response. The customer wakes Alexa either by using a wake word or by pressing the Action button, and then asks Alexa a question that results in a TTS response, such as the current time, a different information response, or a Smart Home request such as turn on the light.
Expected result
The weather forecast is paused during the Listening, Thinking, and Speaking states. When the current time response is completed, the device returns to an Idle state and does not resume or restart the weather forecast.
Alexa is reading the 7-day forecast, or any other TTS response. The customer wakes Alexa, either by using a wake word or by pressing the Action button, and then asks Alexa a question that results in a media response, such as playing music or reading an Audible book.
Expected result
The weather forecast is paused during the Listening, Thinking, and Speaking states. The weather forecast is replaced by the requested media. When the media completes playing, the weather forecast does not resume or restart.
Alexa is reading the 7-day forecast, or another TTS response. The customer wakes Alexa, either by using a wake word or by pressing the Action button, and then asks Alexa to change the device volume.
Expected result
The weather forecast is paused during the Listening and Thinking states. The device volume is changed based on the customer's request, the device returns to Idle state, and the weather forecast doesn't resume or restart.
Note: The weather forecast may restart or resume on products running older versions of the AVS client.
Alexa is reading the 7-day forecast, or another TTS response. The customer wakes Alexa, either by using a wake word or by pressing the an Action button, but doesn't say anything within the stream's timeout window.
or
Alexa is reading the 7-day forecast, or another TTS response. Alexa wakes but cloud-based verification identifies a false wake word.
Expected result
The weather forecast is paused while device is in Listening and Thinking states. When no customer utterance is identified, the device returns to Idle state and the weather forecast doesn't resume or restart.
Note: The weather forecast may restart or resume on products running older versions of the AVS client.
When Alexa is Playing Long Running Media
Alexa is playing long running media such as an Audible book. The customer wakes Alexa, either by using a wake word or by pressing the Action button, and then asks Alexa a question that results in a TTS response.
Expected result
The media is paused while device is in Listening, Thinking and Speaking states. The media resumes once the Alexa response is completed.
Alexa is playing long running media such as an Audible book. The customer wakes Alexa, either by using a wake word or by pressing the Action button, and then asks Alexa for different media such as music or their Flash Briefing.
Expected result
The media is paused while device is in Listening, Thinking and Speaking states. The media is replaced by the new media response, and the original media does not resume or restart when the new media finishes playing.
Alexa is playing long running media such as an Audible book. The customer wakes Alexa, either by using a wake word or by pressing the Action button, and then asks Alexa to change the device volume.
Expected result
The long running media is paused while device is in Listening and Thinking states. The device volume is changed based on the customer's request, and the media resumes playing at the new volume.
Alexa is playing long running media. The customer wakes Alexa, either by using a wake word or by clicking on an Action button, but does not say anything within the stream's timeout window.
or
Alexa is reading media. Alexa wakes but cloud-based verification identifies a false wake word.
Expected result
The long running media is paused while device is in Listening and Thinking states. The device does not go into the Speaking state, and instead resumes the media.
Customer interrupting Alerts
When An Alert is Playing in Short mode
The short mode for an Alert occurs only when the Alert is playing while an Alexa TTS response is also being read. The expected results are the same as for customers interrupting Alexa: when Alexa is speaking, the Alert continues playing in short mode through the entire interaction.
When An Alert is Playing in full mode
An Alert, whether an Alarm, Timer, Named Timer, or Reminder, is playing in full mode. The customer presses the Action button.
Expected result
The Alert is immediately stopped and Alexa is not woken. The device returns to its previous state:
If media was paused when the Alert started, the media restarts or resumes.
If no activity was happening when the Alert started, the device returns to the Idle state.
An Alert, whether an Alarm, Timer, Named Timer or Reminder, is playing. The customer wakes Alexa by voice and uses a voice command to stop the Alert.
Expected result
The Alert goes into short mode while the device is in Listening and Thinking states. The Alert is then immediately stopped but Alexa is not woken. The device returns to its previous state:
If media was paused when the Alert started, the media restarts or resumes.
If no activity was happening when the Alert started, the device returns to the Idle state.
An Alert, whether an Alarm, Timer, Named Timer or Reminder, is playing. The customer wakes Alexa by voice and then asks Alexa a question that results in a TTS response, such as the current time.
Expected result
The Alert goes into short mode while the device is in Listening, Thinking, and Speaking states. The short Alert sound and visual display are played every 10 seconds until the Alexa response completes. Once the response is complete, the full Alert sound and visual display restarts.
Alerts interrupting Alexa
Alerts, such as alarms and reminders, can interrupt Alexa interactions and responses. Each Alert type has a different looping pattern and sound file.
When Alexa is speaking
Alexa is reading the 7-day forecast, or another TTS response, when a Timer or Alarm comes due.
Expected result
The short Timer or Alarm sound and visual display are played every 10 seconds until the Alexa response completes. Once the response is complete, the full Timer or Alarm sound and visual display are played on a loop for 1 hour or until the customer stops the Timer or Alarm.
Alexa is reading the 7-day forecast, or another TTS response, when a Named Timer comes due.
Expected result
The short Timer sound and visual display are played every 10 seconds until the Alexa response completes. Once the response is complete, the full Named Timer audio (sound and TTS) and visual display are played on a loop for 1 hour or until the customer stops the Named Timer.
Alexa is reading the 7-day forecast, or another TTS response, when a Reminder comes due.
Expected result
The short Reminder sound and visual display are played every 10 seconds until the Alexa response completes. Once the response is complete, the full Reminder audio (sound and TTS) and visual display are played on a loop or until the customer stops the Reminder. The looping behavior is defined in the Alexa Reminder directive, currently 2 times.
When Alexa is playing long running media
Alexa is playing long running media when a Timer or Alarm comes due.
Expected result
The media pauses and the full Timer or Alarm sound and visuals are played on a loop for 1 hour or until the customer stops the Timer or Alarm. The media resumes playing once the Timer or Alarm stops.
Alexa is playing long running media when a Named Timer comes due.
Expected result
The media pauses and the full Named Timer audio (sound and TTS) and visual display are played on a loop for 1 hour or until the customer stops the Named Timer. The media resumes playing once the Named Timer stops.
Alexa is playing long running media when a Reminder comes due.
Expected result
The media pauses and the full Reminder audio (sound and TTS) and visual display are played on a loop or until the customer stops the Reminder. The looping behavior is defined in the Alexa Reminder directive, currently 2 times. The media resumes playing once the Reminder stops.
Notifications interrupting Alexa
Alexa is reading the 7-day forecast, or any other TTS response. A new Notification arrives.
Expected result
The New Notification sound plays immediately. The TTS is not paused.
Alexa is playing long running media. A new Notification arrives.
Expected result
The New Notification sound plays immediately. The non-mixable media is not paused.
Alexa is playing a full Alert. A new Notification arrives.
Expected result
The New Notification sound plays immediately. The Alert does not go into short mode.