Since you can use IPA and SSML in Alexa responses, allow IPA in requests for intent utterances and slot values/synonyms.

SSML that uses International Phonetic Alphabet (IPA) in response:

<speak>
I am from <phoneme alphabet="ipa" ph="ʃɨˈkɑːɡoʊ">Chicago</phoneme>.
</speak>
Proposed: Allow IPA in request in either intents or slots:

{
"name": "SAMPLE_City",
"values": [
{
"name": {
"value": "/ˈpɪtsbɜːrɡ/"
}
},
{
"name": {
"value": "Chicago",
"synonyms": [
"/ʃɪˈkɑːɡoʊ/"
]
}
},
{
"name": {
"value": "Phoenix",
"synonyms": [
"/ˈfiːnɪks/"
]
}
}
]
}

This feature request comes off the back of https://forums.developer.amazon.com/questions/200304/phonic-recognition.html

The most recognised and adopted method of teaching kids and grownups to learn english is by using phonics.

Could Alexa also recognise phonics as well as words? This would allow a whole new type of interactive skill that could teach children to read by getting them to practice phonics.

Reference from the UK government for teaching kids to read with sounds:

https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/190599/Letters_and_Sounds_-_DFES-00281-2007.pdf

WakeUp event is a great addition but it could be 10 times better. Right now, you can only send it asynchronously to the Event Gateway, which requires authentication with user's Amazon account. This means more work to get it functional and also more costs for the developers.

If we could just send the WakeUp event in response to the TurnOn/TurnOff directive, directly back to the device that initiated the directive, it would make things more straight forward.

Allow a skill when activated to be interactive with Yes or No commands

Example:

Routine is activated by spoken command

User: "Alexa, I'm home."
Alexa: "Hello, would you like to review your schedule and reminders for today?"
User: "Yes"
Alexa: "You have three events remaining... Would you like to hear your reminders?"
User: "No"

Similar to lex, it would be help to know whether a slot should be elicited prior to the request entering the fulfillment service. I've seen that even if a slot is required, the presence of a slot value must be checked in the lambda function and the ElicitSlot directive returned rather than interaction model determining that ahead of time.

When attempting to build skill, the operation fails due to the skill exceeding 2MB in size. Vote to increase hard limit for size of interaction model!

Right now, it's completely random how AMAZON.Date understands whether day of the week are in the past or the future. ("What's the weather Sunday?")

Half the time AMAZON.Date return last Sunday, the other half the next Sunday.

This should be regimented. Past days should only be returned when the user say "last Sunday" or "this past Sunday", etc.

Really shouldn't need to argue for this. It's completely crazy that this hasn't be dealt with by now!

would like alexa to accomodate not english words for english use in invocation words . An example would be like "Déjà vu." While Alexa is able to interpret this exact pronunciation and spelling when transcribing (speech to text), it is not able to interpret for invocation words because these non-standard characters are not allowed (é, à).

A great feature would be to revision control the interaction model schema and also tool too compare it with previous versions and show history. We work as a team and another great integration would be some sort of github or stash integration .

AFAIU dialogs follow the slot sequence. It would be nice to have the slot sequence numbers in the request for Elicit/ConfirmSlot customization.

Currently slot name/type combination has to be unique per skill. Please make it unique per intent.

Sometimes I want to ask the user for a list of things that are of the same slot type.

For example:
Skill: At what times do you take your medications?
User: 7 AM, 12 PM, and 7 PM

Right now, I have to ask for them individually.

When you want to use your employer's brand name as your invocation name, that name is often not a normal English word. This means that the Alexa voice service often can't understand it and the skill can't be triggered.

A workaround is to write a nonsense word that the Alexa voice service can parse as something that sounds like the brand name, but this is not guaranteed to work.

It would be incredibly useful if the invocation name could contain phonetic markup like responses can.

Should be able to test "Voice & Tone" without having interaction model properly built.

Provide full user utterance for every conversation whether Alexa recognizes the intents/slots or not. There are many cases where the Alexa NLU is getting it wrong and in our fulfillment code we could do a double check and greatly improve the user experience. But we can't double check if you won't tell us what the user says every time.

As mentioned in forums and stackoverflow multiple times. I'm shocked that I couldn't find this when searching feature requests to date.

I am new to Alexa development. I realised that some of my custom type values can alter at runtime. As opposed to doing more NLP processing in the backend, perhaps there should be an API call to allow custom slot type values to be altered on the fly?