Best Practices for Bulbul v3

This page covers Bulbul v3 only (model="bulbul:v3" or omit model for the API default). For the latest persona-based model, see the Best Practice Guide for Bulbul v4 Flash.

Want to hear v3 speakers? See Voices — Bulbul v3.

Choosing the right API mode

Bulbul v3 supports REST and WebSocket streaming. Picking the right transport has a large impact on latency.

	REST API	WebSocket streaming
Endpoint	`POST /text-to-speech`	`wss://api.sarvam.ai/text-to-speech/ws`
Character limit	2,500 chars/call	2,500 chars/session
Latency	Higher — full audio returned at once	Low — chunks streamed in real time
Best for	Short/pre-known text, batch, notifications	Conversational agents, IVR, LLM voice output
Output	Complete audio file	Incremental chunks for progressive playback

For voice agent pipelines (LLM → TTS), prefer HTTP streaming or WebSocket. REST adds dead air while the full clip is generated.

Pace and temperature

Two parameters control how Bulbul v3 sounds. Start with defaults, then tune for your use case.

Pace (0.5 – 2.0)

pace is speaking rate relative to the voice’s natural speed. Default 1.0.

Pace	Effect	Use case
0.8 – 0.9	Relaxed	EdTech, wellness, tutorials
1.0	Natural (default)	Conversational agents
1.1 – 1.2	Brisk	Notifications, IVR, news

Temperature (0.01 – 1.0)

temperature controls expressiveness. Default 0.6.

Temperature	Character	Use case
0.3 – 0.5	Controlled	IVR, BFSI notifications
0.6	Balanced	General purpose, agents
0.7 – 0.8	Warm, conversational	Personas, storytelling

Bulbul v3 does not support pitch or loudness (those apply to bulbul:v2 and bulbul:v4-flash).

Speaker selection by language

Use short-name speakers (shubh, priya, …) with bulbul:v3. Do not use v4 persona IDs with v3.

Language	Code	Recommended male	Recommended female
English	`en-IN`	ratan	ishita
Hindi	`hi-IN`	shubh, ashutosh	priya, suhani
Telugu	`te-IN`	shubh, ratan	neha, priya
Kannada	`kn-IN`	shubh, ratan	neha, ishita
Bengali	`bn-IN`	rehan	roopa, suhani
Tamil	`ta-IN`	ratan, rohan	ishita, ritu
Odia	`od-IN`	shubh	ritu, pooja
Malayalam	`ml-IN`	shubh	pooja
Marathi	`mr-IN`	ratan	priya, ritu
Punjabi	`pa-IN`	mani	roopa, suhani
Gujarati	`gu-IN`	ratan	priya, ritu

Varun is a dramatic character voice — not a neutral default. Reserve for thriller or suspense content.

Use-case presets (v3)

Use case	Language	Speaker(s)	Pace	Temperature	Sample rate
Voice agent	hi-IN	priya / shubh	1.0	0.6	16 kHz
IVR / telephony	hi-IN, en-IN	ratan / ishita	1.1	0.4	8 kHz
EdTech	hi-IN, ta-IN	shubh / ishita	0.9	0.6	22 kHz
BFSI notification	hi-IN, en-IN	ashutosh / ratan	1.1	0.3	22 kHz

Output formats (v3)

Format	Best for
`wav`	Agents, archival
`mp3`	Web and mobile
`linear16`	Real-time agent buffers
`mulaw`	PSTN / 8 kHz telephony

For v3-specific limits (SSML, sample rates on streaming), see the Bulbul model page.