Alexa-Stuff

As a game developer, I've experimented with TTS (Text to Speech) systems such as Festival to produce game dialogue.

Achieving natural-sounding results is a lot of work. Good voice actors are expensive, and for good reason. A lot of work is required. TTS systems are not as much work, but the results are not anywhere near as realistic either (generally placeholder quality only). For instance, while it is possible to get better results, the following link is an indication of the average quality of a Festival TTS Voice.

Alexa is impressive. While the regional voices vary in quality, they all sound fairly natural - with little of the robotic quality often found with TTS systems.

Human–computer interaction (once referred to as HCI) is a very old research topic. Likewise, interactive voice response or IVR has been on the technology horizon for quite some time too. With Alexa and other similiar products the promise of all of this research has finally reached maturity (which doesn't mean to say that further improvement is not possible - in fact, I am sure all of these products will continue to evolve and improve). What has finally been delivered is a fully-fledged VUI (Voice User Interface) with no keys to press or buttons to push; simply talk to it! Or in other words, a nice hands-free option - which is great for preventing distracted drivers.

You can check out my first published Alexa Skill (Word Scramble) here:

http://www.amazon.com/dp/B078WGVWL2

Or my latest Skill (Peanut Allergy Facts):

http://www.amazon.com/dp/B07GB4BG4K

Also in French (Faits sur l'allergie aux arachides):

http://www.amazon.fr/dp/B07GB4BG4K

And now in Spanish (Hechos de la alergia al maní):

http://www.amazon.es/dp/B07GB4BG4K

NLP
Hardware
So what's it good for?
Bits and bobs of stuff for Alexa development
Publishing, operations and the competition
UX for Voice
Privacy
Ethics
Alexa in the media
Icons
Python
To Do

NLP

If you're here because you are interested in natural language processing (NLP), may I suggest you check out my Speech Recognition repo instead? Here I am mostly focussed on Alexa devices and being an Alexa developer. It is possible to use an Alexa SDK for NLP but I have no knowledge of this, sorry.

Hardware

Some general notes on actually using an Alexa device. If you haven't got a Wi-Fi router, you will need one - preferably one that does G/N. And if you are planning on using the web interface, you will need a Wi-Fi card (to talk to the Alexa device).

Router configuration

Set your router to use WPA or WPA2 (the personal edition; I don't believe Alexa supports the professional edition as yet. And WPA2 is probably the better option here) but most definitely not WPA/WPA2. If you are really having problems or just want to get started quickly (please re-configure later) use None.

Set the cipher type to AES unless you are using None (but probably even then too).

Amazon could really have done a much better job on this; by default I think most routers are set to negotiate WPA/WPA2 and TKIP/AES - and Alexa devices will not work if your router is set this way.

App versus Web interface

I was pleased to find that - while an app is probably the way to go - it is entirely possible to set up and configure your Alexa device via the web interface (you will need an Amazon account, also a Wi-Fi card) available at:

http://echo.amazon.com

[This link will redirect.]

As long as you have Wi-Fi enabled, the Alexa device will do a good job of connecting.

You will need to configure the Alexa device to use a permanent Wi-Fi connection; once you have set it up (and registered it to your account) it will download some new software that will allow it to use Bluetooth or pair to a device.

Wake Word

In the phrase:

"Alexa, tell me something"

The word Alexa is what is called a wake word, in other words the trigger for your device.

To change it to something else, either open up the Alexa app on your device and change it there or else head to echo.amazon.com, sign in (you will need an Amazon account), and use the Settings panel to change it.

The choices are currently Alexa, Amazon, Echo and Computer.

Volume

Alexa devices have + and - buttons to adjust the volume.

Of course, this can just as easily be done via voice:

"Alexa, volume 6"

And:

"Alexa, what volume are you set at?"

Also, there are louder and quieter commands too.

Languages

Alexa devices have always supported English but now (since March 2018) support other languages:

http://developer.amazon.com/docs/custom-skills/develop-skills-in-multiple-languages.html

However, developing for languages other than English still seems a little problematic.

Language configuration is at the Alexa device level, so it is possible to have different Alexa devices converse in different languages - for instance English AND French (but it will perhaps be a good idea to have them respond to different wake words so as to save any confusion).

French

For example, the French deployment instructions. On the other hand, from the standpoint of quality, Alexa's French voice is far better than any of Alexa's English voices.

When you submit your skill for certification, the French copy will be carefully vetted for typos and grammatical errors. Any errors will result in a certification failure.

Check out the French version of my Peanut Allergy Facts skill (Faits sur l'allergie aux arachides):

http://www.amazon.fr/dp/B07GB4BG4K

Interestingly, for addressing Alexa devices Amazon prefer the familiar tu form over the more formal vous form. Using vous forms in either the launch phrases or the skill copy will result in the skill being rejected.

There can be some trivial certification failures, which seem to be Amazon protecting its brand. For instance, they prefer the word "skill" to "compétence" (which is probably more correct).

UPDATE: As of October 10, 2018 Amazon announced support for French-Canadian French (fr-CA as opposed to fr-FR). The developer instructions indicate that this may involve using a new i18n module.

FURTHER UPDATE: As of March 21, 2019 Amazon have released a new voice model for French-Canadian French. It seems that there are some substantial differences that required training a unique voice model, as reported by the CBC. In the email I received from Amazon on March 22nd, they refer to "Canadian-French" - but the required language setting remains French (Canada). In my opinion the French-Canadian voice and model far exceed any of the English equivalents (and so are on par with the original French voice and model) but your mileage may vary.

Spanish

Alexa has voices for Spanish, both for es-ES (Spain) and es-MX (Mexico).

As with French, Amazon prefer the familiar tú form over the more formal usted form. Use of the formal forms in either the launch phrases or the skill copy will result in the skill being rejected.

To my ear, neither of these Spanish voices is of the same quality as the French voices, but that may be a personal thing.

Check out the Spanish version of my Peanut Allergy Facts skill (Hechos de la alergia al maní):

http://www.amazon.es/dp/B07GB4BG4K

So what's it good for?

It's a little too early for me to say, but if you're single (Honey, I had a dream you were talking to some woman named Alexa ?!?!) it probably makes a good bedside radio/alarm clock.

UPDATE: As of September, 2018 there is talk of a whisper recognition mode (as far as I know, SSML has always supported a whisper mode for Alexa utterances).

http://developer.amazon.com/blogs/alexa/post/c0e7798d-32bc-4549-9c24-97d204a7bf3a/whisper-to-alexa-and-she-ll-whisper-back

(The U.S. English version will be available in October.)

Most of the following can also be done from a mobile phone, which is only slightly more inconvenient than speaking to an Alexa device. Using an Alexa device is almost always a hands-free operation - which is generally a little more convenient. On the other hand, Alexa will only respond to a wake word whereas a mobile phone can be very susceptible to room noise and chatter (but does not need a wake word).

Alarms

For instance:

"Alexa, set an alarm for 5 minutes"

"Five minutes, starting now"

[Five minutes passes.]

[Alexa devices chimes.]

"Alexa, cancel the alarm." [Or you could press the wake button.]

Timers and alarms can be named, and you can have up to 100 of them.

Apparently they do not depend on the cloud either, being device-resident.

To hear what alarms are set:

"Alexa, what alarms have been set?"

["No alarm has been set."]

As a reference

For those hard-to-spell words:

"Alexa, how do you spell Wikipedia?"

[She also knows weights and measures too.]

As a sleep aid

If you are having trouble sleeping:

"Alexa, open Rain Sounds"

[There are far too many of these to be specific, but you can probably find one you like.]

If you pair Alexa to your smartphone, she can dial numbers for you (but your smartphone could probably already do this).

Wikipedia

Perhaps my favourite skill:

"Alexa, Wikipedia: Hanoi"

[She gives the first paragraph or so; you can ask for more if you wish.]

Bits and bobs of stuff for Alexa development

Now follows some random stuff loosely oriented around being an Alexa developer.

Alexa-hosted Skills

Amazon now offers a beta version of Alexa-hosted Skills which can greatly simplify the development and deployment process. It doesn't even require an AWS account! [But it is limited to the free tier options and limitations.]

Caveats: only Node.js version 8 is supported at present; and session data has changed (if I read things aright) from DynamoDB persistence to S3 persistence.

In my opinion Amazon has greatly simplified the initial coding experience, making Alexa Skills a great deal more accessible to beginners. What is not clear to me is how easy it might be to migrate test code into a production environment.

For my purposes I will be sticking with my established methods - and will not be writing Alexa-hosted Skills.

[UPDATE: there is a "Promote to live" button for Alexa-hosted Skills which I missed. It may very well be that Amazon handles the test -> production migration, although I still need to check this out.]

The Amazon (and Alexa) developer experience continues to evolve by leaps and bounds. While I have not always been a fan of some of their "improvements" I have to say that I am a fan of Lambda Layers (released November 2018), which you can read about here:

http://aws.amazon.com/about-aws/whats-new/2018/11/aws-lambda-now-supports-custom-runtimes-and-layers/

You can find the developer documentation here:

http://docs.aws.amazon.com/lambda/latest/dg/configuration-layers.html

I am currently considering updating all of my Node.js functions from Node.js 6 to Node.js 8, so the following feature may well save me some work:

Lambda Layers are a new type of artifact that can contain arbitrary code and data, and may be referenced by zero, one, or more functions at the same time. Lambda functions in a serverless application typically share common dependencies such as SDKs, frameworks, and now runtimes. With layers, you can centrally manage common components across multiple functions enabling better code reuse. To use layers, you simply put your common code in a zip file, and upload it to Lambda as a layer. You then configure your functions to reference it.

As this will enable me to provision and/or update a single Node.js runtime, I really think this is a very welcome improvement over the current situation (of having separate runtimes for each Lambda function).

As it will also enable AWS console editing of the source code of my Lambda functions, I think this is a big win.

Considering the hazy documentation about Alexa-hosted Skills (which are still a beta offering) I think separate back-end Lambda functions (along with perhaps a common Lambda Layer) will continue to be my preferred arctitecture.

Node.js 8

Although there are many language options in terms of writing Lambda functions, in my opinion Node.js is possibly the best choice at present - both for example code and support from the AWS Lambda functions dashboard.

Amazon announced Node.js version 8 on April 2nd, 2018:

http://aws.amazon.com/blogs/compute/node-js-8-10-runtime-now-available-in-aws-lambda/

One problem with coding in the Cloud is the necessity to upgrade software according to a Cloud Provider's timetable. For instance, I received an email from Amazon on March 24th 2019 warning me that Node.js version 6 would be declared End-of-Life (EOL) on April 2019 (this can be as simple as selecting a different Node.js runtime). [I also received a follow-up email on April 16th. Kudos to Amazon.]

As Node.js has the largest attack surface and greatest number of exploits of any modern language, this is actually good news. The roll-out is not too bad either: slightly more than 30 days (actually 30 days plus a week) warning of the stopping of Node.js 6 activation (April 30) and then 30 days more before Node.js is fully shelved (May 30).

[The above only applies to AWS Lambda functions.]

Amazon provides some background on why upgrading to Node.js version 8 is a good idea:

http://aws.amazon.com/blogs/compute/node-js-8-10-runtime-now-available-in-aws-lambda/

[Well worth a read. TL;DR Node.js 8 features async/await which really simplifies promises.]

Trivia

There are Easter Eggs in Alexa:

"Alexa, cheer me up."

"Alexa, tell me a story."

"Alexa, what's on your mind?"

"Alexa, tell me something."

For instance, according to Alexa:

The word 'boredom' was first coined by Charles Dickens in his 1852 novel Bleak House.

So in a way, boredom was invented by Charles Dickens. This may not come as a surprise to anyone who has ever been required to read one of his very lengthy works!

[This is not a knock on Dickens: I have actually read Bleak House - and I wasn't required to do so, so I guess I was reading for pleasure. My point is that actually requiring anyone to read anything is probably a good way to make sure they don't enjoy the experience. A better route to go is to ban the book; our library had a week of banned books once - and I was very surprised to find that I had aleady read at least half of them. Of Mice and Men, The Adventures of Tom Sawyer, Adventures of Huckleberry Finn or The Catcher in the Rye, anyone? Recently I very much enjoyed The Curious Incident of the Dog in the Night-Time (recommended by a librarian). All, at one time or another, banned. It seems that pretty much everything that is worth reading has been banned at one time or another.]

Glossary

As is usually the case, probably the best place to start is by defining terms or establishing a vocabulary (sometimes called a Domain-Specific Language or DSL). For instance, rather than the over-used app, Alexa developers write skills (but some of the AWS documentation stills discusses skills as if they were apps). The Alexa glossary is actually pretty good (also brief) and well worth reading:

http://developer.amazon.com/docs/ask-overviews/alexa-skills-kit-glossary.html

[As is usually the case, things change quickly in the cloud and the documentation does not always keep pace. For instance, as I write this, Golang is not listed in this Glossary as an option for AWS Lambda functions - and Amazon recently added support for Go to Lambda functions. Caveat emptor.]

For a broader view of things (i.e. not just AWS), check out:

http://www.witlingo.com/voice-first-glossary-of-terms/

Types of Skills

There are various skill types. The main focus here so far has been Custom Skills, although Flash Briefings (which allow for RSS or JSON feeds) also look pretty interesting. Although getting these to pause or stop seems to be a little tricky.

CanFulfillIntentRequest

In the case where a user does not know the specific name for a skill, this new beta offering allows Amazon to select a set of skills that might fulfill the request:

http://developer.amazon.com/docs/custom-skills/understand-name-free-interaction-for-custom-skills.html

Currently in beta (as of August 2018), and US-only.

SessionEndedRequest

This indicates an abnormal end of the current session, most probably a user timeout.

For more information:

http://developer.amazon.com/docs/custom-skills/handle-requests-sent-by-alexa.html#sessionendedrequest

Nota bene:

Your service cannot send back a response to a SessionEndedRequest.

For even more information:

http://developer.amazon.com/docs/custom-skills/request-types-reference.html#sessionendedrequest

[Provides details of the various reasons why the session ended.]

Alexa Intents

While Alexa provides some very useful Standard Built-in Intents it may be a good idea to extend them with Standard Utterances as follows:

AMAZON.RepeatIntent please repeat
AMAZON.YesIntent okay

If the Audio Player option is checked, AMAZON.PauseIntent and AMAZON.ResumeIntent must be specified. Note that these refer to streaming audio, not normal audio such as attending or playback - which will both function correctly without the Audio Player option specified.

For a good example of how not to implement an audio-playing interface, try:

"Alexa, tell me something inspiring"

[Neither "Pause" nor "Resume" seem to work correctly.]

UPDATE: As of April, 2018 this skill seems to have been withdrawn.

"Sorry, I don't know that one."

[This seems to be what she says whenever she doesn't understand your question.]

Alexa Slots

It's a good idea to check up on the defined Alexa slots when considering a voice interaction. Using a predefined slot may well save both you and your users some frustration while getting to grips with using Alexa. The defined slots are mainly oriented to the Amazon shopping experience, but some of them may well be useful to Alexa developers. They add new List Slot Types from time to time (such as AMAZON.Animal), so it's worth checking back periodically. Be careful not to use any slot types marked as Public beta.

While Alexa has some very useful built-in slots, it's probably a 'best practice' to define custom slots. For one thing, the built-in slots are extensive and it is probably a good idea to restrict these quite a bit. Follow the following link for some useful tips on formatting slot values.

Memory Usage

The smallest possible allocation for a Lambda function is 128 MB. So it is a good idea to use this as a target, rather than trying to minimize memory usage with procedural code.

Testing

Rigorous manual testing is of course a first step, but what about automated testing?

Amazon provides a good summary of the various testing options: Why testing and automation matter.

One interesting quote from the above article:

we may not be aware of any problems until a user writes a one-star review.

The Bespoken testing framework is well worth a look:

http://docs.bespoken.io/en/latest/getting_started/

Alexa:

http://docs.bespoken.io/en/latest/tutorials/tutorial_alexa_unit_testing/

SDK:

http://github.com/bespoken/virtual-device-sdk

Example:

http://github.com/bespoken/GuessThePrice

And:

http://bespoken.io/blog/automated-testing-for-alexa-skills-2/

For monitoring your skills in production, logging (see the next section) is essential.

Amazon Cloudwatch alerts can be set to go off if any unexpected circumstances or events arise.

Logging

Each invocation (request or reponse) of a Lambda function will generate billing details, consisting of three CloudWatch events (a START event, an END event, and a REPORT event). These may be filtered out of the CloudWatch console by typing:

-START -END -REPORT

into the Filter events text box (sadly these must be typed in each and every time; a better way to do this is to use CloudWatch Logs Insights and then create a custom Cloudwatch Dashboard).

Amazon published a useful blog post on November 27, 2018 introducing Cloudwatch Logs Insights:

http://aws.amazon.com/blogs/aws/new-amazon-cloudwatch-logs-insights-fast-interactive-log-analytics/

[Note that Cloudwatch Logs Insights will incur usage charges based upon the amount of log data.]

Amazon provides useful query examples, both in the Cloudwatch Insights console as well as at:

http://docs.aws.amazon.com/AmazonCloudWatch/latest/logs/CWL_QuerySyntax-examples.html

The following code can be used to filter out billing details:

fields @message |
filter @type != "REPORT"
 and @type != "START"
 and @type != "END"

The following code can be used to only show exceptions:

fields @message |
filter @message like /Exception/

In my experience CloudWatch Logs Insights includes deleted log entries.

Due to the voluminous billing details, the urge to log each and every interesting user interaction is probably to be avoided. Even so, for debugging reasons it is important to log every significant event (user responses, unhandled events, enough life cycle events for context). There is a rich market for logging analysis, which should be a warning to be careful about what gets logged - for privacy reasons, if not just to reduce clutter.

Of course, it is entirely possible to create a simple skill (such as a Facts skill) that does not require debugging, in which case the performance statistics available from the AWS Lambda console may be sufficient.

Account Linking

This refers to linking to an external (i.e. non-Amazon) account. It looks like standard OAuth.

Permissions

This refers to linking to an Amazon account. There are two groups: location (self-explanatory) and lists. Alexa customers have two default lists: to-do and shopping. Access to these seems to be standard REST. There are two permissions: read and write. Requests are throttled. Follow the link for the Permissions API.

Endpoints

For the purposes of providing content distribution, the region mappings are as follows:

Alexa Developer Console	Region	AWS internal region	Shows in AWS Console as
Default	US West (Oregon)	us-west-2	Oregon
North America	US East (N. Virginia)	us-east-1	N. Virginia
Europe and India	EU (Ireland)	eu-west-1	Ireland
Far East	Asia-Pacific (Tokyo)	ap-northeast-1	Tokyo

[The Endpoints panel on the Alexa Developer Console provides popups that show recommendations. You are limited to regions that provide Lambda functions (initially only North Virginia) - here I have chosen to make my default region Oregon (i.e. on the West Coast) for content distribution purposes but North Virginia might be a better choice. UPDATE: With all of the endpoints specified the default endpoint does not seem to ever be invoked - or at least that has been my experience.]

Bear in mind that any needed ancillary resources (S3, DynamoDB tables) must also be duplicated to all appropriate regions. For databases, the thorny issue of whether or not to replicate the databases across regions needs to be considered (for DynamoDB this might be as simple as using a DynamoDB Global Table).

Interestingly enough, IAM permissions are the exception as they are Global (i.e. apply to all regions). Following security best practices such as the Principle of least privilege, this means provisioning and allocating separate IAM policies and roles for each region.

[Cloudwatch Dashboards are also Global, so that once defined they are available in all regions.]

With some jiggery-pokery, the Endpoints panel can be used to test each and every Lambda function (endpoint), but the Alexa model must be saved & rebuilt every time the endpoints are changed.

Routing

Language variations appear to be routed as follows:

Language	variant	Routed to
English	en-AU	N. Virginia / Tokyo
English	en-CA	N. Virginia
English	en-GB	N. Virginia / Ireland
English	en-IN	N. Virginia / Ireland
English	en-US	N. Virginia
French	fr-CA	N. Virginia
French	fr-FR	Ireland
Spanish	es-MX	N. Virginia
Spanish	es-SP	Ireland

[Sorted alphabetically]

Publishing, operations and the competition

Certification

Once your Alexa skill is complete, you can submit it for certification. Amazon posts a certification results ETA on their developer console.

It's worth checking the certification requirements very carefully before submitting a skill. Amazon will cross-reference these if they reject your skill submission:

http://developer.amazon.com/docs/custom-skills/voice-interface-and-user-experience-testing-for-a-custom-skill.html

They can - and apparently will - reject a skill for fairly arbitrary reasons, for instance if it might compete with their Audible.com business. And they seem to be prone to changing the rules to suit themselves - at the moment, advertising is a no-no, but once they figure out how to commercialize (no pun intended) it I am sure it will again become an option.

Even so, their rejection emails are very specific and actually quite helpful - they highlight the rejected sentences and also provide copy that they would prefer. It is worth noting that the name of your skill must be consistent in all usages - even if this might conflict with normal grammatical usage (think of this as "branding").

Note that certification testing for compliance is not the same thing as logic testing, which is a developer responsibility. It's entirely possible to have a skill certified that still has some bugs in it. There is also a Skills Beta Testing program (with 500 invites) which looks worthwhile.

Certification has gotten lengthier and more stringent over the years (which is a good thing). It took eight calendar days to get my first Alexa skill certified. That skill, resubmitted later with a few minor improvements, only took a day or so to get re-certified. A skill submitted April 12, 2019 was initially scheduled for an April 30th notification - so 18 days, or 12 business days (almost twice as long as a few years ago - but also featured 9 language variations). The initial rejection came April 15 (so about four calendar days) - and took issue with some French copy that had previously passed muster (along with some other things). As VUIs (Voice User Interfaces) continue to gain acceptance, it essential that the quality of the user experience continues to improve. [After some back and forth, the skill was in the end certified on April 24, 2019 - however, this was a pre-existing skill to which I was mainly adding support for Spanish.]

Monitoring and Versioning

Once your skill has been submitted for certification it will be locked on the developer console. However, if you are using AWS Lambda functions, you can monitor the certification testing via Cloudwatch Logs.

Once your skill has been certified (or re-certified), a new development version will automatically be created for you.

While it may be tempting to re-use any existing Lambda functions, it is probably a better idea to create a new Lambda function for each iteration (version) of your Alexa skill.

The Alexa developer dashboard is excellent for monitoring purposes.

Commercialization

In-Skill Purchasing is now generally available.

[Even though I was invited to join the beta, I still haven't looked into this so cannot comment.]

Another development to watch is Alexa Traffic Analysis.

[I may have mis-read this development. Alexa.com proudly proclaims itself an Amazon company, but it may not actually have anything to do with Alexa devices.]

Good demographic data for Alexa is hard to come by, this is all I could find:

http://www.slideshare.net/stevelsilver/amazon-echo-alexa-demographics-75193964

[Published on Apr 19, 2017]

Swag

I got an email March 8th, 2018 that my skill had had more than 100 unique customers within 30 days of publication and therefore qualified for the January Echo Dot promotion. My free Alexa Dot showed up about a half-hour later (now that's shipping!).

Still no custom developer hoodie though. UPDATE: I got an email March 29th that my custom hoodie had been shipped. And it eventually showed up April 2nd (it was addressed to ALEXA DEVELOPER MARTIN RAMSHAW which I thought was a nice touch).

Alexa Dot on the left, custom hoodie in the background. If you do not have a Wi-Fi router (right, not free) you will need one.

UPDATE: As of April 2019 it seems that Amazon have largely dropped free developer swag, although they still feature occasional competitions. Google seem to be playing catch-up however, as they offer t-shirts as well as Google Home devices for qualifying new Actions. [Both companies offer free credits, although once again Google's offering seems to be more generous.]

Alexa versus Google’s Assistant

[I have also looked into Google Assistant, although not yet as deeply as Alexa.]

The first distinction to make when comparing the offerings from Amazon and Google is to differentiate between the software (i.e. the capabilities of the devices - such as Skills or Actions) and the hardware (meaning the devices themselves, such as the Alexa Dot / Alexa Echo / Alexa Echo Plus / Alexa Show - and the Google Home Mini / Google Home / Google Home Hub).

While the devices largely offer much the same features (different designer colours, similiar form factors and sound quality), differing slightly on price and aesthetics, the two offerings are somewhat different in terms of their software capabilities, but not so much that - in this highly-charged competitive market - either offering has a significant advantage. And the situation can only be expected to evolve (and probably rapidly, given that it's a huge market). In short, any comparisons between them must be taken with a grain of salt, given that the situation is very fluid.

Even so, follow the link for an interesting read about Amazon’s Alexa vs. Google’s Assistant.

tl;dr Information is contextual and nuanced, and relying verbatim on either of these devices is problematic. Note that the article is slightly dated, being from 2017.

As of the time of this writing (April 2019) Amazon is apparently leading Google with 15,000+ skills, but as the Google Assistant features a completely different integration model and marketplace this is probably not significant. [Google actively curates its content, so that it's actions do not need to be actively installed - unlike Alexa skills, which need to be individually enabled.]

There are extrinsic factors as well; Alexa can tie into the whole Amazon buying experience (as in: "Alexa, where are my shipments?") - and for entertainment purposes can also tie into Amazon Prime - while Google can integrate with the entire Google platform (as in: "What's on my calendar?" - which ties into Google Calendar).

To see available Google Actions, check out:

http://assistant.google.com/explore

[However, be aware that these actions may not be available in all languages or all regions.]

The devices roughly correspond as follows:

Amazon	Google
Alexa Dot	Google Home Mini
Alexa Echo	Google Home
Alexa Echo Plus	no Google equivalent (as far as I know)
Alexa Show	Google Home Hub

The smallest devices are fine for voice interaction, but not so great for listening to music (although newer Alexa Dot devices have an audio-out jack). The mid-range devices feature better speakers (and so are better for listening to music), while the high-end devices also have display surfaces.

UX for Voice

While Amazon provides quite a lot of content related to Alexa, sometimes an external source may provide a broader perspective. Some of the following may be of interest:

http://www.lynda.com/User-Experience-tutorials/UX-Voice-Planning-Implementation/679644-2.html

http://softwareengineeringdaily.com/2018/05/18/alexa-voice-design-with-paul-cutsinger/

Note that UX for Voice is not IVR, which is a much older technology.

Privacy

There are some troubling privacy aspects to having voice devices in your household or place of work. Here is an interesting video from Vox about some of the implications:

http://www.youtube.com/watch?v=IUEN2TAqlmU

[The video was posted on Dec 28, 2020.]

Now that adoption of voice-activated devices has become widespread the situation can only be expected to evolve.

One or two interesting quotes follow.

medium.com

From a medium.com article:

We expect to enjoy privacy in our own homes, but this is ultimately incompatible with the reality that the data collected by your Echo belongs to Amazon, and not you.

And:

So while virtual assistants themselves are not inherently problematic, they do bring to the fore many of the problems that have built up around our use of centralised data-driven services, and especially so for home Internet of Things devices.

Both from:

http://medium.com/oxford-university/whos-storing-your-conversations-e5736fc5d9f5

bloomberg.com

From a bloomberg.com article:

Amazon.com Inc. employs thousands of people around the world to help improve the Alexa digital assistant powering its line of Echo speakers. The team listens to voice recordings captured in Echo owners’ homes and offices. The recordings are transcribed, annotated and then fed back into the software as part of an effort to eliminate gaps in Alexa’s understanding of human speech and help it better respond to commands.

From:

https://www.bloomberg.com/news/articles/2019-04-10/is-anyone-listening-to-you-on-alexa-a-global-team-reviews-audio

The article is well worth a read as it gives quite a lot of background on Amazon and their voice training practices.

wired.com

It is routine to have human experts review audio snippets to ensure accurate STT transcription.

This is not without some troubling aspects:

Google paused human audio review worldwide in July after reports that a contractor was leaking audio snippets in Dutch.

Despite the headline of the article, this is not just a Google issue.

Apple:

About three weeks after Google paused audio snippet review, Apple did as well on August 2.

Amazon:

Amazon ... offers the option for users to opt-out of sharing audio clips for transcription. The company launched a dedicated "Alexa Privacy Hub" in May and added the voice command, "Alexa, delete everything I said today," soon after.

Microsoft:

Microsoft said at the end of August that it no longer uses human review for audio snippets

Facebook:

Facebook said that it would stop human review on audio from Messenger

And:

All the major smart assistant developers maintain that audio snippets are fully anonymized before they go out to reviewers and most specifically say that less than one percent of total interactions with smart assistants actually get reviewed by a person.

All from:

http://www.wired.com/story/google-assistant-human-transcription-privacy/

[Wired specializes in stories about technology that are not overly technical. Well worth a read.]

Ethics

For techies and tech companies, the ability to be able to do something has usually out-weighed ethical considerations.

[This has been a long-term trend, with science and technology generally out-pacing legal and ethical considerations. However, legislation can be relied upon to eventually catch up to technology. Rather than having to expensively retrofit privacy safeguards into an existing product strategy, it behooves innovators to behave proactively and consider legal and ethical concerns from the outset. This can be an effective differentation strategy, and very possibly an effective means of securing a competitive advantage. The end result might very possibly be a corporate privacy policy. And then of course there is the troubling issue of affinity groups.]

While legal considerations concern themselves with the law (and can be considered to be the minimum requirement), ethical considerations go above and beyond and are generally concerned with morality. Just because something is legal does not necessarily mean it is moral. While legal penalties can include fines and prohibitions, ethical penalties can include loss of consumer confidence, goodwill, market share and revenue.

For those interested in such things, the contrasting ethical frameworks are Deontology and Utilitarianism. Deontology is the idea that an action is either right or wrong, and can be assessed on that basis. For instance, respecting a customer's right to privacy. Utilitarianism is what is known as a consequentialist theory - in that actions can be evaluated in terms of their consequences - and can broadly be categorized as the end justifies the means. For instance, the idea that the benefits of aggregating a properly-anonymized collection of user utterances outweighs the costs of a number of violations of customer privacy (these things generally seem to be framed in terms of cost-benefit analysis).

Just as ethics is a growing concern in the machine learning world, so it is also becoming important in the voice world, and not simply from a privacy perspective either. As the following Fortune article should make clear, failing to take into account ethical considerations can be expensive - if not in a purely legal sense, then in a brand, market-share or professional prestige way:

http://fortune.com/2018/05/11/google-duplex-virtual-assistant-ethical-issues-ai-machine-learning/

As prominent sociologist Zeynep Tufekci put it: “Google Assistant making calls pretending to be human not only without disclosing that it’s a bot, but adding ‘ummm’ and ‘aaah’ to deceive the human on the other end with the room cheering it… horrifying. Silicon Valley is ethically lost, rudderless and has not learned a thing.”

[The above quote is from the preceding Fortune article.]

Alexa in the media

You know Alexa is mainstream when Saturday Night Live parodies it:

http://www.youtube.com/watch?v=YvT_gqs5ETk

Here is a UK advertisement:

http://www.youtube.com/watch?v=sulDcHJzcB4

Alexa-Stuff

Contents

NLP

Hardware

Router configuration

App versus Web interface

Wake Word

Volume

Languages

French

Spanish

So what's it good for?

Alarms

As a reference

As a sleep aid

Wikipedia

Bits and bobs of stuff for Alexa development

Alexa-hosted Skills

Node.js 8

Trivia

Glossary

Types of Skills

CanFulfillIntentRequest

SessionEndedRequest

Alexa Intents

Alexa Slots

Memory Usage

Testing

Logging

Account Linking

Permissions

Endpoints

Routing

Publishing, operations and the competition

Certification

Monitoring and Versioning

Commercialization

Swag

Alexa versus Google’s Assistant

UX for Voice

Privacy

medium.com

bloomberg.com

wired.com

Ethics

Alexa in the media

Icons

Python

To Do

About

Languages