Skip to main content

1. What is Sarvam TTS?

Sarvam TTS is a high-performance text-to-speech service built specifically for Indian languages. Its Bulbul model family delivers natural, expressive synthesis tuned for conversational use cases such as voice agents, IVRs, and customer support bots, with real-time streaming over both HTTP and WebSocket.

2. Key features of Sarvam TTS

Indic-First Voices: 10 Indian languages plus English (India), with voices trained on native speakers rather than accented English voices reading Indian text. Two Voice Catalogs: 222 Bulbul v4 Flash voices, each built for one language and use case, plus 16 curated Bulbul v3 speakers that work in every language. Low Latency Streaming: Designed for real-time use, so Bolna can start playing audio before the full response has been synthesized. Pace Control: Speaking rate is adjustable per agent, exposed in Bolna as the Speed rate slider.

3. How Bolna uses Sarvam for TTS

Bolna integrates Sarvam TTS to power Indian-language voice agents across recruitment, sales, and support workflows — generating prompts, questions, and responses in native languages for better engagement, especially in Tier 2/3 regions. Real-Time Speech for Seamless Conversations: Sarvam’s streaming synthesis lets Bolna agents speak as the LLM produces text, keeping the conversation flowing instead of pausing between turns. Multilingual & Accent-Aware Voice Support: Each agent — or each language inside a multilingual agent — gets a native voice, so a Hindi turn is spoken by a Hindi voice and a Tamil turn by a Tamil one. Handling Complex Pronunciations and Technical Terms: From candidate names to role-specific jargon, Sarvam TTS pronounces Indian names and domain terms correctly where English-first models mangle them.

4. List of Sarvam TTS models supported on Bolna AI

Sarvam has retired bulbul:v1, and bulbul:v2, Sarvam’s previous generation with speakers such as anushka and abhilash, is no longer selectable on Bolna. Agents still pointing at either must move to bulbul:v4-flash or bulbul:v3. Voice ids do not carry across model generations: a v3 voice fails on v4 Flash and a v4 Flash voice fails on v3, so changing the model means picking a new voice from the lists below.
A voice from an older Bulbul generation fails agent creation with 400 Provided voice: <voice> is not available for the provider: sarvam.

5. Voices

Bulbul v4 Flash

Bulbul v4 Flash has 222 voices. Each voice id names a persona, a language and a use case as <persona>_<language>_<use case>, for example aparna_hi_customer, simran_en_sales or kavitha_te_conversation. Many personas come in several styles, such as sanchita_hi_banking, sanchita_hi_kyc and sanchita_hi_edtech. A v4 Flash voice is listed only under its own language. Malayalam and Odia have no v4 Flash voices, so agents in those languages use bulbul:v3. voice_id is the full id and voice is its display name, such as Aparna (Customer). Browse them in the Languages tab or list them with List voices.

Bulbul v3

Bolna offers 16 Bulbul v3 voices. The same set is available for every supported language, so pick by gender and tone rather than by language. voice_id is the lowercase id from the first column; voice is the display name from the second. Both are required in provider_config.
The catalog changes over time and can differ if you have connected your own Sarvam key. Confirm the current list with List voices, or browse and preview them in the Languages tab.

6. Supported languages

Bulbul supports 11 languages — 10 Indian languages plus English (India). Send the plain code in provider_config.language; Bolna resolves it to the locale Sarvam expects.
Sarvam’s speech-to-text and text-to-speech models do not cover the same ground. Saaras STT reaches 22 Indian languages plus English on Sarvam’s API; Bulbul TTS covers the 11 below. A language your agent can transcribe is not necessarily one it can speak.
On bulbul:v4-flash the spoken language follows the script of the text, not language. A reply written in Tamil script is spoken in Tamil even when language is hi, so keep the agent’s prompt and its v4 Flash voice in the same language.
language inside provider_config is required for Sarvam. Leaving it out fails with 400 Tools Config > Voice > Language: This field is required, even when the agent’s transcriber already has a language set.
Assamese (as) is accepted by Bolna’s transcriber and passes synthesizer validation at agent creation, but it is not one of the 11 languages Bulbul supports — Sarvam’s TTS API has no as-IN. Place a test call before shipping an Assamese agent on Sarvam TTS, or pair Sarvam transcription with a synthesizer that covers Assamese.

7. Configuration

voice is the display name and voice_id the lowercase id, and both are required. speed maps to Sarvam’s pace control, is the Speed rate slider in the dashboard, and accepts 0.5 to 2.0 on both models. Bulbul v3’s other parameters, temperature and dict_id for custom pronunciation dictionaries, are not exposed through Bolna. (pitch and loudness existed only on the older bulbul:v2, which Bolna no longer offers.) See the Create Agent API for every synthesizer field.

Conclusion

Sarvam TTS brings localized voice synthesis to conversational AI in India. By integrating Bulbul, Bolna ensures its voice agents are not only intelligent but also relatable and linguistically inclusive — improving candidate experience, response rates, and accessibility across diverse demographics. For related integrations: