How to train your smart speaker | radioinfo

How to train your smart speaker

Thursday 30 May, 2019

Analysis from Steve Ahern.
 
 
 
 
Smart speakers are not quite as smart as we are led to believe.
 
I have no doubt that coming generations of smart speakers will deliver huge amounts of voice driven artificial intelligence powered interactivity that will benefit the radio and audio industries, but we’re not there yet.
 
The current generation of smart speakers are actually pretty dumb. They rely on tightly controlled routines and they try to train us rather than us training them.
 
Don’t get me wrong. I love my smart speakers and can see the potential, but I can also see a range of improvements needed to take them to the next level, as intelligent audio assistants and smart media interfaces.
 
They are becoming an important delivery platform for live radio streaming. One network told me they are now seeing about 60% of their online stream listening via smart speakers.
 
Over the past few months I have been training my smart speakers to customize them to my needs, but I have also been aware that they are training me to their way of doing things.
 
I have an Echo dot, enabled with Commercial Radio Australia’s RadioApp, and a Google Home mini, which I take with me when travelling.
 
I logged into my Google speaker with one of our radioinfo gmail accounts, but I didn't already have an Amazon account so I created one to log into Alexa. I have linked my free basic Spotify, Amazon and Youtube accounts - none of them are paid premium services.
 
Here is what I have found in several months of experimentation with these devices and the improvements I would like to see as the devices evolve.
 
Setup
 
Setup of both types of speakers requires a linked phone, downloading of the controller apps and a connection to wifi. Setup is not always easy or straightforward. None of the setup routine is voice activated. Once plugged in and powered up, the speakers sit there waiting to be linked to something. If you speak to them at this stage they politely answer that you need to connect them to an app and to wifi.
 
Here is improvement number one - it would be a good idea if the speakers immediately scanned nearby wifi networks and smart phones upon activation, then, instead of waiting for instructions from you, they prompted you to complete each step by asking if you want to connect to your home network, to your phone, to load the app and for passwords.
 
Listening to the radio
 
My Amazon Alexa speaker came with instructions on how to initialize the speaker and link it to RadioApp. The manual setup took about 15 minutes, including the time to download the Alexa app, to create an Amazon account and to download the Alexa Assistant app.
 
I immediately started talking to the Alexa speaker to see what it could do. Because I had loaded and linked the RadioApp, it could quickly find all the Australian commercial and ABC radio stations that I asked for. This was a quick win which gave me instant gratification for the basic functions of the Alexa speaker. But there was still more to be done to take it to the next level.
 
My Google Home took slightly more time to set up, despite that fact that I already had a Google account and I already had Google Home installed on my smart phone. This speaker is currently not compatible with RadioApp, so it was not able to deliver me the same instant gratification for finding Australian stations that I experienced with Alexa.
 
With Google, when I asked  ‘OK Google play Gold FM,’ I got a random station from some other country played to me via TuneIn, rather than Gold 104 Melbourne, which I wanted. When I refined my search by saying ‘play Gold FM Melbourne,’ it did eventually find the station from TuneIn.
 

Customisation
 
Once your smart speaker has been enabled, its time for basic training.
 
I tried to train my speakers to learn a range of my favourite stations. I was not very successful. They did not remember my regular stations and which feed they came from, and instead they trained me to ask more specifically for what I wanted by saying things like ‘OK Google, from iHeartRadio, play KIIS 106.5.’
 
The most significant example of how the devices trained me is related to location and callsign. I like to listen to Gold FM stations on both the Gold Coast and in Melbourne. When I was in Sydney requesting Gold FM, Google Home could not distinguish my preference and it did not remember the last Gold FM played. When I was in Melbourne, it knew my location (I turned location services on) and played me Gold FM Melbourne.
 
If I asked Google for the old callsign of ABC Radio Adelaide (ABC891) it didn't find the station from its default service, TuneIn. If I asked it specifically to find the station from another streaming source, such as iHeartRadio, then it was able to play it for me.
 
If I said ‘Play Gold FM Melbourne/GoldCoast,’ or if I used the FM call numbers, it found the correct station. It had trained me to be more specific in how I requested the station. By the way, one very interesting thing that Gold FM Gold Coast does is to begin its feed with the latest local news bulletin, that’s a really good idea, however, sometimes the live feed does not follow the bulletin.
 
Asking Alexa for commercial or ABC stations was much easier. RadioApp knew most of them and has been trained to recognize a variety of callsigns. Alexa knew all the old and new callsigns for ABC Radio, and for rebranded stations such as 2UE (now Macquarie Sports Radio), because CRA has worked hard to figure out the many possible ways that someone could ask for a station name and programmed that info into RadioApp.
 
There are some anomalies though.
 
Alexa has adopted all the new callsigns for SCA stations, but doesn’t recognize the SCA legacy callsigns. When I asked for B105 I got a country music station in Cincinnati called B105.1 via TuneIn. When I asked for Hit 105, it gave me the correct station. Same result for 4TO and other old SCA callsigns across the country. There is probably nothing to be lost by SCA programming the old callsigns into the back end of the RadioApp system so that listeners who still refer to those stations by the old names will be able to find them on their smart speakers.
 
ABC News
 
Asking for ‘ABC News’ generated the greatest anomaly.  There are two elements to examining ‘ABC News.’
 
First, both Google and Alexa allow you to set up routines to enable you to do common tasks. These routines can play music from your streaming account, get your favourite podcasts, and choose news services from a range of providers.
 
Australian ABC Radio has provided a news service to Google and Alexa for inclusion in these routines (also known as Flash Briefings). ABC America, the BBC, New York Times and many others also provide audio news services to incorporate in your routines.
 
I have set up my ‘Good morning’ routine to tell me the weather and traffic then play three news services, ABC News (Australia), BBC News and the New York Times.
 
The anomaly is that the ABC News feed comes from ABC TV (I can tell by the news theme), and not radio. When listening overnight the feed is sometimes out of date if there is a breaking international story, whereas news from ABC Radio’s hourly bulletins is always live and has the most up to date information. ABC seems to not be using the news bulletin audio from its ABC Listen app, but sourcing it from somewhere else to feed this service.
 
The second anomaly is that ABC Radio’s news channel is difficult to find. Because it is called ‘ABC News,’ when I ask for ‘ABC News’ the speakers tell me to go to my flash briefing and play me the tv news bulletin, they don’t play me the ABC News radio station that is available on air and on the ABC Listen app. There may be a case for ABC Radio to reconsider the naming of its radio news channel or at least distinguishing it from the other services that the national broadcaster feeds to RadioApp, TuneIn, iHeartRadio and other audio feed services that supply smart speakers.
 
Naming and branding is an issue in the world of smart speakers, but the solution is easy, put as many relevant names, brands and callsigns as possible into the SSML meta data so that the artificial intelligence engine can find the station you want. This metadata is invisible to people and only accessed by other machines, so there is no chance of confusing the audience, it is simply a way to help them find the station they want to listen to.
 
Feeds are another thing for radio companies to think about. Do they want to supply their feeds to third party aggregators such as TuneIn, which feeds most smart speakers, or do they want to control their own feeds. This is a subject for further discussion within the industry as these new devices develop.
 
Ad replacement
 
The only station I noticed doing ad replacement during my testing was Triple M. I was listening simultaneously to Triple M Melbourne on both my Google and Alexa speakers and noticed something interesting. Both feeds were pretty much in synch with eachother. Google was picking up the feed from TuneIn and Alexa from RadioApp, both feeds were identical until the ad break. My location was set to Sydney on both devices. Alexa played me ads for Sydney products during the ad break, while Google (via TuneIn) played me the Melbourne ads that were being broadcast live. Both feeds ended with a station ID then went to a song. The Alexa feed returned to the live stream a fraction of a second late, but most people would not notice that. Nice work SCA.
 
Community Radio
 
Community Radio is well behind the rest of the Australian industry when it comes to smart speakers. Community stations are not on the RadioApp or iHeartRadio feeds.
 
Some community stations are listed on TuneIn, but many of their audio streams are poor quality or silent. The community sector has a lot to do to catch up in this area.
 
Beyond the Basics

Nova has done the most to build interactivity into its Alexa smart speaker feeds. You can ask Nova what song is playing and a range of other requests that bring a deeper level of interactivity to radio and will be a big part of the future.

One thing though, the artificial intelligence behind the Nova Alexa skill is not intelligent enough to know that I am already listening to Nova, so to trigger the correct response I always have to preced my request with 'Ask Nova...' otherwise my smart speaker does not know which skill routine to activate to give me the answer. I should be able just to say 'Alexa, what song is that,' rather than remembering to say 'Alexa ask Nova what song is that.' This is another improvement needed in the next generation of smart speakers.

 
Like training a dragon, the wrong decisions could leave us with our fingers burnt. Radio industries around the world need to be thinking and discussing all their options about how to maximize advantage in these new devices.
 
Below is a table of the requests I made to my speakers as I tried to train them. You can see the responses from each speaker in the columns.
 
If you have a speaker, try it yourself and give us feedback and further information in the comments box below (or send your comments to editor@radioinfo.com.au).
 
We are at the beginning of an exciting new era, we need to understand it, debate all the implications, and develop options in the best way for our industry. This is a new field for all of us, but it is one that listeners are embracing quickly, so it will become increasingly important for radio stations everywhere.
 

 

About the Author
 

Steve is the founding editor of this website.

He is a former broadcaster, programmer, senior executive and trainer who now runs his own company Ahern Media & Training Pty Ltd.

He is a regular writer and speaker about trends in media. More info here.

 

 

 

Advertisement

 

Post a Comment

0 Comments

Log InYou must be logged in to post comments.
radioinfo ABN: 87 004 005 109  P O Box 6430 North Ryde NSW 2113 Australia.  |  All content © 2012. All Rights Reserved.