Train your voice¶
A voice that is yours, on the same streaming engine, for the same 20 paise a minute. The process is a recording session, a training run with the same quality gates as our stock voices, and a new voice id in your project.
What we need from you¶
- About 30 minutes of clean speech from one speaker: a quiet room, a decent microphone, 44.1 kHz or 48 kHz WAV (or better), no music, no background voices. We send a script and a checklist; reading your own real call scripts in your actual Hindi/English mix gives the best result.
- The language mix the voice will speak: Hinglish, Hindi, Indian English, or another Indian language (ask: coverage is growing).
- Where it will play (IVR, app, agent) and a rough monthly volume, so we can plan capacity.
- Rights: confirmation that the speaker consents to the voice being synthesised for your organisation's use.
Timeline¶
| Step | Who | Time |
|---|---|---|
| Request (portal → Train your voice) | you | 2 minutes |
| Script, checklist, scheduling | us | 1–2 days |
| Recording | you (we can help remotely) | half a day |
| Training + evaluation gates (ASR word-error, listening test) | us | about a week |
| Voice live under its id in your project | us | same day as the gate passes |
Progress shows under Train your voice → Your requests (received → scheduled → recording → training → live).
Using it¶
The voice appears in GET /v1/voices for your keys; keys created with "all voices" can use it immediately, per-voice keys need the new
id added. Everything else, including price, is identical.
Exclusivity and removal¶
A voice trained from your recordings is used only by your organisation. Ask and we delete the weights and the recordings; usage logs (which hold no audio) stay for billing.