Timing metadata
Every rendered file ships with phoneme, word and sentence boundaries. Captions, karaoke style highlighting and IVR menu handoffs slot in without a second pass.
A voice API sized for real product traffic.
Everything the studio does is exposed as REST endpoints, with streaming audio, predictable latency, and IVR ready output formats. The API is the same surface the studio itself renders on.
POST/v1/voice/render
{
"voice": "aanya",
"tone": "warm",
"language": "en-IN",
"script": "Welcome. Your table is ready.",
"format": "wav",
"sample_rate": 24000,
"stream": true
}Audio, plus the metadata a production system actually needs.
Streaming output
Audio can be streamed as it renders. First byte latency stays low enough for voice bots and live IVR without a pre buffer stall.
Formats and sample rates
WAV, MP3, OGG, and PCM at telephony grade rates for IVR and studio grade rates for content. Server side transcoding on request.
Latency and throughput
Median first byte under a second on standard voices. Concurrent render slots scale on Business tier with dedicated capacity.
Three patterns cover most of what you would build.
Batch render
Push a list of scripts, receive a list of audio URLs. Fits content pipelines - explainer videos, audiobooks, marketing spots.
Live streaming
Open a socket, stream text tokens, receive audio chunks in flight. Fits voice bots, live IVR, conversational agents.
Pre rendered assets
For IVR menus and fixed prompts, generate once, serve from a CDN, refresh on script change. Predictable cost, near zero latency.