tl;dr

  • Speaking requires multiple cognitive and physical processes, all at once, and each leaves a detectable trace in your voice.
  • Soma reads three types of signal: how you sound, your timing, and what you say.
  • We test every measure before we ask you to rely on it.
  • Your Soma Score is built from your own baseline and read on a shared scale. An 80 means the same for everyone, but what your 80 sounds like is yours alone.
  • A score tells you what changed in how you spoke. It does not tell you why, and nothing here is a diagnosis.

Why voice

Say something out loud, anything. For most people this feels effortless, yet you just did something amazing.

A seemingly simple sentence can require your brain to search through memories, decide on meaning, find the right words, manage your breath, and coordinate the muscles of your throat and mouth, all in the same instant. Throughout your day, each of those jobs leaves its mark on your voice, and that’s why speech is such a rich signal of both your mind and your body.

This is the core idea that Soma is built on. You speak all the time, projecting your voice out into the world, and Soma measures its hidden signals from the conversations you already have each day.

Fig. 2.0 One sentence broken down into three signals.

What we measure: three types of signal

The information carried in your voice naturally falls into three types.

How you sound

Think of the last time you were nervous before a presentation and heard your own voice come out tight. Or the flat “I’m fine” from a friend who is not fine. Pitch, loudness, rhythm, and the texture of the voice are among the most consistent audible signals of stress and low mood. Thinking hard shows up here too. When people are under mental load, there are small but measurable shifts in the voice.

Your timing

You know how you sound after an all-nighter. Researchers have measured it. Kept awake around the clock, people’s speech changes: the pitch and timing shift, and the melody flattens. Pace and pauses behave very differently, though. How fast you talk is one of the steadiest things about your speech. How you pause is one of the most variable.

What you say

Notice where your “um” lands in a tough conversation. It tends to arrive just before the hard part, while you are still deciding what to say. Soma reads the words themselves: how varied your vocabulary is, how often you reach for fillers like “um” and “like”, and how much you hedge with “maybe” and “sort of”.

No single signal tells the whole story. Together, the three complement each other to give a full picture of how you spoke today, measured against your own normal.

How we know a signal is real

A number from your voice is only useful if you can trust it, so the scientists at Soma test each measure before we ask you to rely on it. Here are two examples of those tests.

1. Does it come back the same?

Step on a bathroom scale twice. If the two numbers are far apart, you have learned more about the scale than about your own weight. Voice is harder to weigh, because you never say the same thing exactly the same way twice.

A lot of speech measurements don’t give you the same answer twice. Reliability has to be demonstrated, not assumed, so we check that. We take many conversations from the same person and ask how much of a measure stays put and how much moves around. The part that stays put is you. The part that moves is the moment. How fast you talk is a great example. Your individual speech rate mostly reflects who you are, not what day it is.

Fig. 3.0 How much of each measure is you, and how much is the moment. Measured on Soma users’ anonymized recordings.

2. Does it match what is really there?

A scale can show the same number repeatedly, yet still be wrong each time. A second test is against something real that we know. For pitch (how high or low your voice sounds), ground truth can be measured using a laryngograph, a sensor on the neck that directly records the vibration of your vocal folds. We’ve tested our pitch readings against laryngograph measurements, and run our own code on public data to see whether it reproduces what’s already known.

Fig. 3.1 A few seconds of one public research recording. The dark line is pitch measured straight from the vocal folds. The amber line is pitch read from the microphone. Where the lines overlap, the microphone reading is right. Contains information from PTDB-TUG (Graz University of Technology), made available under the Open Database License.

These are just two examples. There are more, and we are not done. Every measure in Soma goes through checks like these, and we continue to test as the product grows and improves. We will share what we find here, including the results of our validation studies as they are completed. Science is the foundation of Soma, and we believe you should be able to see it.

Compared with your own normal

You pick up the phone, hear one “hello”, and instantly recognize that it’s your mom. Voices differ enormously from person to person: fast or slow, high or low, full of melody or quite flat. None of that is good or bad. It’s just who you are.

Comparing your voice with everyone else’s mostly tells you what you already know: you sound like you. The more useful comparison is with yourself. In studies where people recorded short voice notes across their days, drifting from their own usual pattern lined up with how they said they were feeling in that moment.

This is where your Soma scores come in. Soma learns your baseline from your conversations: your usual level on each measure, and how much you normally vary around it. A score then shows how far today sits from that baseline on a scale from 0 to 100. That scale is the same for everyone, so an 80 means the same thing for you as for anyone else, even though no two voices sound alike.

Your first baseline is a rough picture. It sharpens as you record more, and it draws on several days, so a single off day doesn’t define it.

Fig. 4.0 After one conversation, comparing you with yourself is no better than comparing you with everyone else. With every conversation you add, your own baseline gets sharper. Measured on Soma users’ anonymized recordings.

What a reading can and cannot tell you

A reading from your voice tells you how you spoke in one conversation, and how that compares with your own baseline. On its own, it does not tell you why. A change can come from sleep, from the room, from who you were talking to, or from any number of other factors.

That is why Soma doesn’t only show the numbers. It can connect to your calendar and your wearable, and set each reading alongside what was happening in that moment. Your calendar shows which meeting you were in and what the day around it looked like. Your wearable adds how you slept and how hard your body was working. Insights point to the moments where your voice shifted and suggest what might be behind them. They are prompts, not conclusions. Soma lets your voice and your context come together to paint the whole picture.

A voice is a rich signal, not a verdict. No single sentence or conversation should be read as one, and nothing here is a diagnosis.

Who we are

Everything we build at Soma starts with the science. We are a growing team of scientists, engineers and designers, and our job is to be the skeptics in the room. Before a number reaches you, someone on the team has asked whether it is real, and searched for that answer in the data.

You have been speaking your whole life, and your voice has been carrying this signal the entire time. We are building Soma so that you can finally hear it.

So say something out loud. There is more in it than you think.