When machines want to judge your tone

Amazon’s new Halo health wearable device will track your physical activity, your sleep patterns — and, by listening to your voice, your emotional tone.

Why it matters: The device’s tone-tracking service raises questions about user privacy, but it’s also part of a growing industry that employs AI and voice-recognition to analyze the emotional affect in a human voice.

How it works: In many ways, the $99.99 Halo is a standard wearable, tracking a user’s health with the help of an accompanying smartphone app.

  • What sets it apart is a small mic on the band that can record snippets of your voice, which is analyzed using machine-learning to take into account pitch, intensity, tempo and rhythm.
  • Those bits of speech are timestamped and identified with labels like «content» or «hesitant,» as well as measured for «positivity» and «energy level.»

Not surprisingly, the idea of a device from one of the biggest tech companies analyzing the emotional tone of a user’s voice raised «Black Mirror» comparisons.

[2027:] «In a statement, police said the suspect’s Halo device detected “anger,” potentially leading credence to the officer’s account of fearing for his life….» https://t.co/7qIUO783XK

— sid thee fussell (@sidneyfussell) August 27, 2020

Yes, but: Users have to opt in to the tone feature, and Amazon emphasized that speech samples are recorded locally on the phone — not shared on the cloud — and are automatically deleted after processing so that no one can listen to them.

Still, what’s known as «sentiment analysis» is increasingly being used by businesses in voice communication, often to help sales agents interact with customers — a need that has grown as agents carry out their work remotely during the pandemic.

  • But even some of those in the field caution that machines lag behind humans when it comes to interpreting emotion. «It’s natural for people to be empathetic,» says Zayd Enam, the co-founder of Cresta, an AI startup that coaches sales agents in real time. «Our goal is to give agents the answers so they can focus on the human component.»

The bottom line: It’s still best to let human beings — not machines — police your tone.

 
Info photo

Amazon’s Halo wearable, with Tone feature enabled on a smartphone. Photo courtesy of Amazon

 
Πηγή: axios.com

Σχετικά Άρθρα