Back to basics - CAI and NLP 003

This week, I'm travelling back to the beautiful Tunbridge Wells, where I lived for 10 years.

As I sit writing this, in a Premier Inn (classy, I know), I'm reminded about the importance of doing the basics well.

I'm not talking about the all you can eat breakfast, or the proximity to the train station, or the price of a room. I'm talking about the fact that at 3am last night, four random, intoxicated guys on their way home from a wedding happened opened our room door with their keycard.

Luckily, I'd literally just been to the loo, so I was awake and rushed to the door to learn that someone at reception had checked them into our room and given them a keycard that would open our door!

This is so basic.

Speaking of basics, voice-enabled elevators seem to be a no-brainer. Who wants to touch those buttons these days? Especially when you don't know who's potentially COVID-riddled hands have also been on there.

And even though we've seen many prototypes of voice enabled elevators, like the now infamous sketch from Burnistoun:

Or the slightly more serious software demo from Picovoice:

Still, voice-enabled elevators haven't quite hit the mainstream, yet.

However, it looks like they might, as Nuance has just signed on the dotted line to bring voice-enabled elevators to "one of the top manufacturers in the world", according to Cerence CEO, Sanjay Dhawan.

New releases

Google GoEmotions

Google releases GoEmotions , a dataset for training emotion classification is AI systems. For conversational AIs to be truly human-like, they need to be able to detect and respond to emotions like a human would. To date there are pretty much no applications I've seen that couple emotion detection with conversational AI. For example, the ability to recognise frustration in a user would enable a bot to hand over to a human quicker.

With GoEmotions, it looks like we're getting there.

It's worth noting that GoEmotions was trained on Reddit data and so is detecting sentiment from text, not from voice.

Jovo v4

We're such a huge fan of Jovo at VUX World. We've built many solutions using the framework and are proud to be listed as preferred partners.

This week, Jovo released its v4 framework which includes the following updates:

• New Debugger : An all-new Debugger with multimodal testing capabilities
• Components : Reusable, isolated elements for a better project structure
• Output : A multimodal template engine that works across platforms
• CLI : Rebuilt from the ground up, including plugins and custom commands
• More : Lots of more improvements and updates

Rasa v3

In the same week, as if by magic, Rasa releases v3 of its open source NLU which includes:

• Architecture revamp
• Slot mapping improvements
• Experimental feature: markers

Partnerships

"Chaucer and Sovran announce strategic partnership to leverage voice-bot technology for human-centric omnichannel transformation to optimise staff performance and boost customer experience."

What does that mean? Well, it pretty much means that Chaucer will likely licence or white label Sovran AI's technology to deliver solutions to its clients.

Many AI vendors are taking this partnership approach because large consultancies have existing trusted relationships with clients, which makes selling-in a new technology easier than the vendor trying to reach and attract the client directly.

Over time, I predict we'll likely see some of the larger consultancies acquire some of these AI startups in order to provide their own technical solutions to clients. Accenture already try this with a bundled AWS packaged up as a conversational platform.

Another reason why brands and enterprises such tread with a little caution when engaging with certain companies, as they may have vested interests in selling you their solution, rather than the right solution for you.

(For the record I'm not saying that this is what Chaucer or Accenture are doing. But something to keep an eye on as you develop your requirements.)

Funding

"Transcription AI startup Verbit has closed a $250 million Series E funding round led by Third Point Ventures. The new funding pushes Verbit’s valuation to more than $2 billion half a year after achieving?$1 billion unicorn status ."

"Nvidia is adding $3.5 million to conversational AI startup Kore.ai’s $70 million Series C funding round and teaming up to integrate voice AI technology for customer service centers."

This deal is specifically to enable Kore.ai to roll out custom brand voices using NVIDIA's Riva technology. Riva can apparently create decent sounding TTS voices with relatively little training data. That means it's less expensive and more accessible for brands.

We'll see custom brand voices gather pace in future as brands realise that the only way you can actively differentiate your voice assistant from your competitors or create meaningful, memorable customer experiences, is to use audio to your advantage. Custom brand voices are one way of doing that.

I, however, am always a little bit skeptical when I hear about realistic human-sounding voices that can be created with minimal training data. The proof will be in the pudding, which I'd like to see.

Acquisitions

The trend of CPaaS providers acquiring conversational Ai capabilities continues. We saw Vonage acquire Over.ai years ago. Genesys acquiring Bold 360, the same. Now, Link Mobility acquires Matelab?Xenioo to add conversational AI capabilities to its CPaaS customers.

The aim for these CPaaS providers is to be a one-stop shop for all communication automation needs and so it's no surprise to see continued acquisitions here.

As with all of these things though, as a brand, your requirements should inform your technology choice, not your existing vendor, so I'd take all this with a pinch of salt.

AudioCodes acquires Callverso

Perhaps more interestingly, is AudioCodes acquisition on Callverso .

AudioCodes Voice AI Connect product is used to integrate bots from any bot framework into any call centre . And one of its core selling points is its agnosticism on both sides. Any bot framework. Any call centre.

This acquisition will give AudioCodes the capability to build bots for its customers (or allow customers to build themselves), integrating those into call centres with Voice AI Connect.

This may well make sense for some customers who are perhaps less mature in their capability and skills than others. However, what this does do, is potentially place AudioCodes in competition with the very bot platforms it’s being trying so hard to partner with.?

Whether this could lead to conflicts of interest or a misalignment of incentives remains to be seen. I'm genuinely a fan of AudioCodes positioning of its Voice AI Gateway and certainly hope that the open spirit remains.

Opinions

Voice search is not the future . That's according to an SEO expert at Google who's built his entire career on traditional SEO and has a personal vested interest in keeping things as they are. I'll say no more.

Can Sonos be the first company to crack the AI privacy game and enable big growth? Find out what I reckon.

Awards

Congratulations to the Webio team for winning the 2021 AI Award for Best Application of AI in a Customer Service Application at the AI Awards Ireland. The Webio team are doing great work in revolutionising debt collections using AI. Hats off.

Knowledge

How we won the Alexa Prize SocialBot Grand Challenge

Jan Sedivy is a legend in the voice AI space, with vast experience working on speech systems at IBM and Google. He recently guided the Czech Technical University team in winning the Alexa Prize SocialBot Grand Challenge 4; a competition where contestants from Universities across the globe aimed to design a conversational AI that’s able to hold a conversation for 20 minutes.

Jan joins us to share the journey through the Alexa Prize, how the team approached the project, lessons learned and conversation design tips geared to increase engagement with AI bots. He also shares some insights from his career in voice tech and his latest venture, Promethist AI.

Tune in this Wednesday by registering below.

Conversation design best practice in the enterprise

Celene Osiecka joined us to share tips and insights on designing conversations at scale, teamwork, personality and talent.

Rewatch the LinkedIn live broadcast below, listen to it on your chosen podcast player or watch it on YouTube .

Finally, should you use agile project management in your AI projects? Here's a simple framework to determine whether you should... Or shouldn't.

That's all folks. Until next time, see you later.

---

VUX World is the front door to the world of AI-powered customer experience, helping business leaders and teams understand why voice, conversational AI and NLP technologies are revolutionising how we live, work and get things done.

We help executives formulate the future of customer experience strategies and guide teams in designing, building and implementing revolutionary products and services built on emerging AI and NLP technologies.

Find out more about how VUX World can help your brand succeed with AI.