Overview
-
Sarvam AI launched Bulbul V4 Flash on October 13, 2026, a real-time text-to-speech model for 11 Indian languages with under 250 ms time-to-first-audio.
-
It costs Rs. 20 per 10,000 characters, one-third less than V3, with a half-price offer of Rs. 10 until November 12, 2026.
-
Developers can start through the Sarvam API, Playground, or Content Studio, and pick REST, HTTP streaming, or WebSocket based on how live the use case is.
Sarvam AI debuted Bulbul V4 Flash on October 13, 2026, aiming to address existing shortcomings in India’s voice technology market, where many speech tools sound stiff in regional languages and experience lag during live calls.
Sarvam, the Indian AI lab, designed this model to converse in 11 languages with natural pacing and generate initial audio in under 250 milliseconds. This guide outlines the model’s capabilities, pricing structure, and the steps developers can take to begin using it.
What Bulbul V4 Flash is and Why Sarvam Built It
Bulbul V4 Flash is a text-to-speech model developed by Sarvam AI that converts written text into spoken audio. It is tailored for real-time applications such as customer support, banking calls, payment reminders, and sales outreach.
The model functions as a compact, distilled version of Sarvam’s broader research initiatives, successfully handling regional accents, Indian English, names, numbers, and mixed-language speech like Hinglish.
Featuring a new text normalization layer, it reads currency figures, acronyms, and identification numbers the way a human would. Sarvam states its objective is “not just speaking India’s languages, but speaking them the way India does.”
Additionally, Sarvam develops the Sarvam AI LLM family—including Sarvam 105B—alongside speech-to-text and vision models, establishing itself as a prominent builder of Indian AI models.
Also read: Sarvam Becomes India’s Newest AI Unicorn With USD 234 Million Funding Led by HCLTech
Speed, Price, and Language Numbers Worth Knowing
According to Sarvam, the model achieves a time-to-first-audio of under 250 milliseconds in production settings and operates up to 40 percent faster than Bulbul V3 under comparable loads.
Pricing is set at Rs. 20 per 10,000 characters—a one-third reduction from V3’s rate of Rs. 30. A promotional launch offer further reduces the price to Rs. 10 until November 12, 2026. Under the promotional rate, processing 1 million characters monthly would cost approximately Rs. 1,000, or Rs. 2,000 at standard pricing.
The model supports 11 languages: Hindi, Bengali, Gujarati, Kannada, Malayalam, Marathi, Odia, Punjabi, Tamil, Telugu, and Indian English. It includes 83 voices and 224 voice-and-style configurations, backed by a speech dataset nearly seven times larger than that used for V3.
Because compact models can blur speech patterns across languages, Sarvam trained V4 entirely from scratch with Indian languages as its primary focus.
How Developers Can Start Using the Sarvam API
Getting started involves four steps. First, developers sign up via the Sarvam dashboard to generate an API key. Next, they select the model ID (`bulbul:v4-flash`), specify the language code, choose a voice, and submit the text. The API then delivers the audio in the requested format.
Non-developers can preview the voices using the Sarvam Playground or Content Studio. Teams utilizing AI coding tools can also incorporate Sarvam’s text-to-speech skill to generate functional code. Available output formats span MP3, WAV, AAC, Opus, FLAC, and telephony options like mu-law and A-law. Furthermore, Sarvam provides migration documentation for teams transitioning from ElevenLabs, Cartesia, Deepgram, or Gemini.
Choosing Between REST, HTTP Streaming, and WebSocket
The REST API is ideal for brief tasks, supporting up to 2,500 characters per request. HTTP streaming accommodates up to 3,500 characters and streams audio dynamically as it is generated. For live voice agents requiring interactive dialogue, WebSockets are recommended.
To minimize latency, Sarvam suggests keeping WebSocket messages under 500 characters. The default sample rate is 22,050 Hz, with an 8,000 Hz option available for phone lines. Developers can also adjust speed, tone, and volume, and integrate a pronunciation dictionary for specific brand names.
Pre-built integrations exist for platforms such as LiveKit, Pipecat, Twilio, Exotel, and Vapi, allowing teams to embed the voice engine into existing call infrastructure.
Also read: India AI Impact Summit 2026: PM Modi Tests Sarvam AI Smartglasses, First Hardware to Launch in May 2026
Best Uses & Limits of this Indian AI Model
Contact centers, financial institutions, and educational platforms stand to benefit the most. Delivering a loan reminder in Tamil or a scheme update in Marathi with a localized accent rather than a robotic tone helps build consumer trust. Because voice quality directly influences credibility in automated calls, natural pacing functions as a critical business consideration.
Certain limitations apply. The model supports 11 out of the 22 official Indian languages, and because speed and quality metrics are provided by Sarvam, organizations should conduct independent testing with their own scripts. Additionally, while Sarvam has previewed Bulbul V4 Pro for enhanced speech capabilities, no release date or pricing details have been announced.
Why this Sarvam AI Voice Model is Worth a Test
The Sarvam AI voice model offers Indian enterprises an expedited and cost-effective method to communicate with customers in regional languages. It supplies 11 languages, 83 voices, a sub-250 millisecond response time, and a pricing structure of Rs. 20 per 10,000 characters.
Conducting small-scale tests using production scripts will demonstrate how accurately the model manages proper nouns, numbers, and Hinglish. For development teams constructing voice applications, this Indian AI model warrants consideration.
FAQs
1. What is Sarvam AI Bulbul V4 Flash?
It is a text-to-speech model created by Sarvam AI that converts written text into natural speech for live applications including customer support, banking interactions, and sales calls.
2. Which languages does it support?
It supports 11 languages: Hindi, Bengali, Gujarati, Kannada, Malayalam, Marathi, Odia, Punjabi, Tamil, Telugu, and Indian English, alongside mixed-language phrasing like Hinglish.
3. How much does it cost?
The standard rate is Rs. 20 per 10,000 characters, with a promotional launch price of Rs. 10 available through November 12, 2026.
4. How can developers start using it?
Developers can generate an API key via the Sarvam dashboard, reference the model ID `bulbul:v4-flash`, configure the language and voice, and transmit text using the REST, HTTP streaming, or WebSocket APIs. Voices can also be evaluated directly within the Playground or Content Studio.
5. Is it better than Bulbul V3?
Sarvam reports that V4 Flash is up to 40 percent faster, one-third less expensive, and delivers superior speech quality. Because these benchmarks originate from Sarvam, teams are encouraged to evaluate the tool using their own materials.
