Voices
Listen to every voice before you pick one for your product. Each sample below reads the same expressive line, “वाह, यह तो कमाल है! Sarvam AI के साथ अपनी भाषा में बात करना अब बहुत आसान हो गया है।” (“Wow, this is amazing! Talking in your own language with Sarvam AI has become so easy now.”), so you can compare voices directly and hear how each one handles natural emotion and prosody. Use the speaker parameter to select any of these in your API calls; see How to change the speaker voice for code samples.
Want to test your own text? Head to dashboard.sarvam.ai to generate live previews with custom input.
Not sure which voice to pick? Bulbul v4 Flash best practices lists expert-ranked personas for English and Hindi by use case.
Need a voice that isn’t in this list? The Instant Voice Cloning API (beta) generates speech in a cloned voice from a short reference clip, or in a saved voice via voice_id — usable from the same SDK alongside these speakers.
Bulbul v3 Male Voices
shubh, not Shubh).varun produces a deep, dramatic tone suited to suspense or villain characters. It is not recommended as a neutral default and should be reserved for thriller, drama, or suspense use cases.
Bulbul v3 Female Voices
Bulbul v4 Flash Voices
bulbul:v4-flash uses the same request and response contract as bulbul:v3: pass model="bulbul:v4-flash" with one of the speaker values below (or omit speaker to use the default persona shubh_enhi_ads). The catalog is organized by persona rather than a flat name list. Most voice IDs follow <name>_<language>_<style> (for example ishita_hi_devotional, simran_en_customer). The middle token is a short language tag (hi, en, bn, …) or enhi for English–Hindi code-mixed personas.
Languages in this catalog (222 voices): Hindi, English, English–Hindi (enhi), Bengali, Gujarati, Kannada, Marathi, Punjabi, Tamil, and Telugu — mapped at request time with the usual BCP-47 language_code values (hi-IN, en-IN, and so on). Bulbul v3 still covers the full 11-language set (including Malayalam and Odia); v4 Flash personas are not offered for every v3 language yet.
Speaker names are case-sensitive and must be lowercase, exactly as shown in the tables below.
Featured voices
Start here if you are unsure which persona to pick. These 11 voices are hand-picked for the most common production use cases. Filter by use case, copy the exact speaker value, or grab a ready-to-run snippet.
shubh_enhi_companionshubh_enhi_bankingshubh_hi_ecommaparna_hi_customershalini_enhi_companionishita_enhi_companionsanchita_en_companionroopa_hi_companionashok_hi_newsanand_hi_documentaryaparna_en_edtechAudio previews for bulbul:v4-flash voices are not available on this page yet. Generate a live preview from dashboard.sarvam.ai or with a quick API call using the speaker value from the cards above or the tables below.
All v4 Flash voices
Hindi (hi-IN) (116 voices)
English (en-IN) (62 voices)
Hindi-English Code-Mixed (13 voices)
Marathi (mr-IN) (9 voices)
Kannada (kn-IN) (4 voices)
Punjabi (pa-IN) (4 voices)
Tamil (ta-IN) (4 voices)
Telugu (te-IN) (4 voices)
Gujarati (gu-IN) (3 voices)
Bengali (bn-IN) (3 voices)
Hear Voices Across Languages
Bulbul v3 supports 11 languages, and voice quality varies by language. Below, two of our top-rated speakers, shubh (male) and ishita (female), read the same localized line in each supported language, translated with Sarvam Translate.
For v4 Flash, start from the use-case voice rankings. For v3 short names, see Bulbul v3 best practices.