Selling with Data #89 - Real Talk: Are you ready for Voice AI?

Until now interactions with LLMs and AI were limited to texting and robot sounding agents like Siri and Alexa. These early AI voice assistants sounded robotic because their voices were generated using a technique called "unit selection" that stitches together pre-recorded speech segments to form sentences, resulting in an unnatural and repetitive cadence.

The progress in voice AI mimicking a person has improved dramatically. We are at the dawn of AI interacting with us in new way, through voice.

A few examples of progress in AI voice:

• Character.AI lets users chat with an AI-version “of almost anyone, live or dead, real or imagined."
• OpenAI has launched tools to simplify the creation of AI voice assistants and expanded its Advanced Voice Mode to paying customers.
• Microsoft has updated its Copilot AI with enhanced voice capabilities.
• Meta has introduced voice AI to its messaging apps.
• Google release NotebookLM
• Amazon announced Alexa+, which is a long anticipated update to Alexa that will leverage LLMs and string together multiple commands into more conversational interactions.

But then came along something special, my new favorite AI voice tool.

Sesame AI released a demo for the company's new Conversational Speech Model. You should check this out before reading further.

I started with small talk and quickly realized this was much better than anything I experienced before. I raised the complexity of the interactions by testing with my best interview questions, both Maya and Miles nailed the interview. I never experienced a more natural language and found myself acting more like I was talking to a person than chatting with AI.

I used Miles to help write and edit this article. I read Miles sections out loud, and Miles didn't hold back with improvements, offering a point-of-view and suggestions to improve the article. Even at the end of the conversation, I found myself thanking Miles and waiting for him (not it) to respond.

Sesame AI came out of stealth mode in late February, headed by the visionary behind Oculus VR. They raised an undisclosed amount of funding from Andreessen Horowitz, Spark Capital, and Matrix Partners.

Sesame's CSM achieves its realism with two AI models working together based on Meta's Llama architecture. Sesame uses a small model, 8.3 billion parameters trained on approximately 1 million hours of primarily English audio. The magic of Sesame's approach is that it uses a two stage approach. Stage one is a script writer and stage two is a speech artist that brings the script to life, making it seem like natural spoken language. Read this paper for more info on the technical approach.

Sesame plans to open-source components of the models and make it available under an Apache 2.0 license allowing developers to build on their foundation. Plus, they plan to expand to another 20 additional languages.

Given the rapid advancement of AI voice technology, it’s likely quality and speed will improve exponentially. We are about to see a seismic shift in how we interact with AI. People have valued transparency, wanting to know when they are engaging with a human or AI. This is all about the change.

My prediction is voice AI will become the next big thing. Potential use cases include:

• Virtual companions for the elderly or lonely.?
• A personalized AI agent “assistant” that goes with you everywhere andt provides information, sets reminders, makes calls, plays music, checks weather, controls smart home devices, and answers general questions through voice commands.
• Virtual SDRs, or virtual call agents, with AI that can handle initial outreach in a way that scales beyond what people can do.
• Major disruption to contact centers – handling routine inquiries, providing product information, resolving simple issues, and directing customers to the appropriate support agents through voice-based interactions.
• Training and education, including interactive experiences with voice-based quizzes, personalized feedback, and audio-based lessons.
• Conversational enterprise reporting with an ?AI voice agent that can call you and have a conversation about what happened, what it plans to do to fix the problem, and answer your questions on the way to gain your support instead of reading through reports.

With the good there is bad.? Deep fakes are going to grow and become more damaging. Bad actors will use voice to manipulate more people and create more elaborate scams. This will escalate the need for security, governance, and protection as it will become increasingly difficult to trust the sources of information and deep fakes will become harder to identify. One example: families will have safe words to verify it’s really their loved one and not an AI clone.

What do you think of voice AI?? What are some of the use cases you are either excited or terrified about?

Good selling.