Voqalize / Company
Our story
We spent two years on the hard part of voice.
We set out to build voice products, and most of the work turned out to be underneath them. Here’s how it turned into Voqalize, and why we’re letting other teams build on it today.
Where we started
We were building products people could talk to.
We started out building voice products, software you could talk to and get something done.
The first versions looked great in a demo, in a quiet room with someone who knew what to say.
Real people
Then real people talked to them.
People don’t talk in clean sentences. They stop to think, change their minds halfway through, leave the TV on and talk from a train.
Each of those breaks a voice product, and none of it is the thing you set out to build.
They pause to think.
The product answers into the pause and cuts them off.
People take turns with gaps of about 200 milliseconds (Stivers et al., PNAS 2009), so a pause to think looks a lot like being done.
They change their mind.
The product keeps talking, then carries on as if they heard all of it.
The TV is on.
The news ends up in the request.
The signal drops.
The answer arrives late, and the conversation falls apart.
Underneath
So we kept building underneath.
We fixed one break at a time, and each fix became another layer under the product.
That’s where the two years went.
- Audio that gets through a bad signalMicrophones, WebRTC, every browser, a weak connection on a train.
- Hearing only the personNoise, echo, and the product’s own voice coming back through the speaker.
- Getting the words rightNames, numbers, addresses, two languages in one sentence.
- Knowing when they’ve finishedA pause to think, or the end of the sentence.
- Stopping when they cut inOn the word, and remembering only what they heard.
- A voice that starts fastSpeech that begins while the reply is still being written.
- A character that moves with itA face whose lips match the voice.
- A record of every conversationRecordings to go back to.
The same list
Every team adding voice was building the same thing.
Along the way we saw that any team putting a voice on its agent would need the same layers. Most start with the same estimate, two weeks, and then meet the same list.
It made more sense to build it once, for everyone.
A stack for others
We turned it into a stack for other teams.
Voqalize is the voice AI infrastructure we built for ourselves, now ready to sit under your product.
Bring your own agent and Voqalize adds the voice. Or let Plug & Play run the agent and the voice for you. Either way, Voqalize handles the listening and the speaking, and your product keeps its own logic and screens.
You bring the brain. We bring the voice.
Today
Today we’re opening the doors.
Most of our work went into listening and timing. Starting today, any team building a product people talk to can use it.
“Every demo we built worked in a quiet room. Then a real customer called with the TV on. Most of what we built since was for that customer, and now other teams get it too.”
Voqalize is built by a team in Bengaluru, India.