A Bharatanatyam teacher runs her whole studio by voice
DanceHall Studios is a studio-management app built for South Asian classical dance schools. CBB Labs built a voice agent for it — Sangs, running on AssemblyAI speech-to-text. With it, one studio now runs close to 4x more of its day-to-day work just by talking, and the owner wins back six hours a week.
A studio action is one real thing getting done: an announcement or email sent, a subscription paused, a calendar event added, course material updated, or feedback sent to students.
What she says, and what happens
Real things she gets done in her week — just by talking to Sangs. Pick one.
She says
“Send reminders to the Tuesday class about the recital rehearsal.”
What happens
Sangs drafts it in her house style and sends it to the whole cohort. She can shape it as she speaks — “make it warmer,” “drop that last line” — and every correction teaches the system how she writes.
Before
~15 min typing it out on the dashboard
With voice
Spoken and sent in seconds
The load around the teaching
Shivakami, director of Temple of Fine Arts, Texas, and a DanceHall Studios user, has been a classical dancer for more than 30 years and has taught for over 10. Her studio has close to 100 students and growing, across 13 courses, and she teaches globally. Everything around the teaching is the load: class emails to every cohort, course setup, uploading practice videos one by one, pausing and resuming a student’s subscription, and running the monthly billing. With no admin sitting next to her, all of it lands on her, and it pulls her away from the students.
From typing to talking
DanceHall’s dashboard gave her visibility into student progress, attendance, and revenue — everything a modern SaaS platform should have. Agentic actions added another layer of convenience, but interacting through chat still meant thinking, typing, and waiting. In many cases it took about the same time as clicking through the dashboard yourself.
Then we added voice agentic actions.
The moment she stopped typing and started speaking, everything changed. Instead of opening dashboards or composing prompts, she simply asked questions and gave instructions: “Who hasn’t paid this month?” “Send reminders to the Tuesday class.” “How many students attended last week?”
The language of the dance
Bharatanatyam is one of the world’s oldest continuously practiced classical dance traditions. Rooted in Tamil Nadu and shaped by centuries of Sanskrit and South Indian musical traditions, its vocabulary blends Tamil and Sanskrit. This is the part that makes speech-to-text genuinely hard. Teachers casually use terms like thattadavu, nattadavu, tattimetti, alarippu, jatiswaram, and varnam. If the transcription gets one of these words wrong, she has to read the text again, edit the mistake, and break her flow. At that point voice stops saving time and starts feeling like another thing to manage.
We tried several speech-to-text providers. AssemblyAI was the one that handled these terms reliably. She can speak a full message in this vocabulary and it comes back right, without stopping to re-correct, and the accuracy keeps improving as our own knowledge base of her terms grows: AssemblyAI does the listening, and our harness learns her world on top of it. From there, everything becomes simple. She speaks, AssemblyAI transcribes, and Sangs turns her words into the right action. So far, about 9 out of 10 requests made by voice carry out the intended action correctly — practical enough for everyday use.
What it gives back
With voice, two things moved: the amount of work and the time it takes. Shivakami now does close to 4× as much studio admin as she did over chat, in about 10% of the time. Sending emails, creating courses, uploading materials, and managing student billing are all done by voice now.
For example: 15 minutes typing an email, 15 minutes creating a new course, and 30 minutes adding links over chat and saying where each one goes — now all of it is done in under 5 minutes. Shivakami saves 6 hours of admin work every week. She no longer operates software; she delegates it through voice, less like using a tool and more like having an assistant.
Why voice adoption was instant
Here is what we think happened. She never had to learn a new habit. She already runs her day by voice, sending her students WhatsApp voice notes the same way she talks to them in class. So when the software started listening, there was no adoption curve to climb; the interface finally matched the behavior she already had.
Chat asked her to translate herself into typing. Voice just let her talk.
The takeaway
Just as CBB Labs added voice agents like Sangs to DanceHall Studios, we think any business that runs on administrative work will move to voice the same way, and faster. For DanceHall, we are extending voice beyond emails, materials, and billing into feedback, sign-ups, and onboarding, and strengthening our evals so the agent stays reliable as it takes on more. The stronger the evals get, the more gets done by voice.
We expect DanceHall Studios to become voice-first within one to three years — customer management, finance, and operations all run by talking.
Figures from DanceHall usage: the ~4× and ~90% are drawn from this owner’s activity logs (text era vs. voice era, sampled); the ~9 in 10 is from a sample of her voice requests; hours saved are her own estimate.
“It feels like Sangs is my chief of staff. I just say what I want, send the email, pause a subscription, set up the calendar, and it’s done.”