Amazon Playing Nice with Brands?

The news last week that Amazon had been working with KFC to create a custom brand voice for their Alexa skill has the brand world excited about reclaiming some ownership in the voice space. As brands look to assert more control over their voice assistant presence, we expect brands with unique characters or spokespeople to be especially keen to embrace this so their experiences are immediately identifiable and distinct from a generic assistant voice. In truth, all brands could benefit, as this recent study from Princeton & University of Chicago showed that consumers have real trouble distinguishing between 1P and 3P experiences. Along with this new capability, Amazon is also courting brands by testing a new Alexa audio ads offering similar to what competitors Spotify and Pandora have launched. While these moves may look brand-friendly, Amazon has a history of fractured relationships with brands who view the company’s scale as both opportunity and threat.

Header image source

A Voice of Their Own

Amazon is now opening up their AWS Amazon Polly service to assist brands in creating text-to-speech voices. Similar to how the Samuel L. Jackson voice was created for Alexa, the service can take a sample of voice recordings from an individual and then deploy their language processing model to create an entire synthetic voice. KFC has leveraged this technology already to create a Colonel Sanders voice for their Alexa skill. More on The Verge.

Conversationalizing Documentation

As healthcare IT leader Cerner prepares for the HIMSS20 conference, they have been vocal about their goals to leverage voice technology and improve patient and care provider services. According to a Cerner intelligence executive, “The new approach of conversational voice, or ambient voice as some call it, brings the promise of getting the same data from a natural interaction with a patient and does not require data entry from the care team.” In the near term, the company will be focusing on using voice recognition to improve data entry and documentation services. Long-term they hope to be able to scale the service and mine its data to help provide insights to caregivers and identify and enact improvements to operations. More on Healthcare IT News.

Fool Me Once…

MIT has developed a NLP testing software, called TextFooler, which has successfully “fooled” leading NLP systems used with virtual assistants including Alexa, Google Assistant and Siri. The software “looks for the words in a sentence that are most important to an NLP classifier and replaces them with a synonym that a human would find natural.” By doing so, TextFooler was able to make the AIs mis-understand the intent of sentence variations whose meaning would be nearly identical to a human. The hope is that tools like TextFooler can be used to uncover weaknesses in these NLP models and point them toward areas in need of improvement. More on MIT Technology Review.

Stats Don’t Lie

10.8% projected amount of all digital buyers in the US that will make a purchase using a smart speaker this year. (data via eMarketer)

7,554 Google Actions available in Hindi for Assistant (vs. 768 Alexa skills available in Hindi) (data via Voicebot AI)

Chart of the Week

eMarketer US Smart Speaker Users 2017-2021
(data via eMarketer)