A voice for what
you’re building.
Bring mathematical narration into learning tools, publishing workflows, and ideas we haven’t imagined yet.
01. Prepare a speech plan
Submit typed text and LaTeX segments. The plan exposes the prepared reading, source correspondence, unresolved issues, and a credit quote. A spoken_override lets you give an explicit intended reading when notation needs context.
curl "$EVARIA_API_URL/v1/plans" \
-H "Authorization: Bearer $EVARIA_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"domain": "mathematics",
"language": "en",
"segments": [
{"kind": "text", "text": "Consider the following integral."},
{
"kind": "latex",
"text": "\\int_0^1 x^2\\,dx",
"spoken_override": "the integral from zero to one of x squared, with respect to x"
}
]
}'Inspect the returned plan before requesting audio. The example supplies the integral’s reading explicitly; it does not demonstrate automatic equation interpretation.
02. Inspect, then generate
Use the returned plan ID and exact quoted credits. Supply a unique idempotency key for each intended job; reuse that key when retrying the same submission. Credits are reserved once.
curl "$EVARIA_API_URL/v1/jobs" \
-H "Authorization: Bearer $EVARIA_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: $UNIQUE_REQUEST_ID" \
-d "{\"plan_id\":\"$PLAN_ID\",\"expected_credits\":$QUOTED_CREDITS}"03. Listen and review
Poll the job, then download and review the audio when it reaches review_required. In the trusted engineering preview, accepting the entire recording charges its reserved credits; rejecting it releases them. This manual review workflow is still being evaluated for a public product.
| METHOD & ROUTE | PROTOTYPE BEHAVIOR |
|---|---|
POST /v1/plans | Prepare a reading and exact credit quote. |
GET /v1/plans/{id} | Retrieve an immutable speech plan. |
POST /v1/jobs | Queue synthesis with idempotent credit reservation. |
GET /v1/jobs/{id} | Check generation and review state. |
GET /v1/jobs/{id}/audio | Download authorized job audio when available. |
POST /v1/jobs/{id}/review | Accept or reject after listening to the entire audio. |
GET /v1/credits | Inspect available, reserved, and charged credits. |
Predictable units, visible preparation
The prototype counts one credit per Unicode code point in prepared spoken strings, including internal spaces and punctuation. Explicit pauses and internal retries add no credits. The spoken expansion of an equation can be longer than its LaTeX input.
The current planning contract supports 1–80 segments, up to 1,500 characters per text segment and 15,000 total content characters. These are engineering limits, not a production service commitment.
Built with a clear first focus
Our launch focus is English mathematics. We are working toward meaningful specialist evaluation before making claims about correctness, performance, latency, or supported scale. Tell us the types of material, integrations, and review controls that would make Evaria useful to you.