The new text-to-speech model is designed for low-latency voice generation, with potential applications across conversational AI, customer support and multilingual digital services.
Bengaluru-based artificial intelligence company Sarvam AI has introduced Bulbul V4 Flash, a text-to-speech model designed to generate real-time voice responses across 11 languages, including 10 Indian languages and Indian English. The model targets applications such as conversational AI, automated customer support and enterprise voice agents.
According to the company, Bulbul V4 Flash delivers a time-to-first-audio latency of under 250 milliseconds, allowing applications to begin generating spoken responses with minimal delay. The model is intended for services where response speed is important to maintaining natural interactions between users and AI systems.
The latest release supports Hindi, Bengali, Gujarati, Kannada, Malayalam, Marathi, Odia, Punjabi, Tamil, Telugu and Indian English. It is designed to address pronunciation, pacing and contextual delivery across languages commonly used in India.
Sarvam AI said its research focused on improving the naturalness of speech generation while reducing the time and cost associated with deploying voice-based applications.
The company has priced Bulbul V4 Flash at ₹20 per 10,000 characters, compared with ₹30 per 10,000 characters for its predecessor, Bulbul V3. The revised pricing represents a reduction of approximately 33%, making the new model less expensive for developers and enterprises processing large volumes of text.
The model also introduces an expanded catalogue of more than 200 persona-based voice options, organised according to language, speaking style and application requirements.
Developers can access Bulbul V4 Flash through application programming interfaces (APIs), including REST, HTTP streaming and WebSocket connections. These integration options allow businesses to incorporate speech generation into existing applications and digital workflows.
The model can be used in interactive voice response systems, virtual assistants, educational platforms, automated service interactions and audio content generation.
For customer-facing applications, lower latency can help reduce pauses between a user's query and an AI-generated spoken response. This is particularly relevant for voice agents handling customer enquiries, appointment scheduling and other automated interactions.
Sarvam AI has positioned its speech technologies around India's multilingual requirements, where users frequently switch between regional languages and English during everyday conversations.
Its developer guidance also addresses language-specific pronunciation, number formatting and the handling of mixed-language text, which can affect the quality of generated speech.
Bulbul V4 Flash builds on the company's existing speech AI portfolio, which includes transcription, translation and text-to-speech capabilities for Indian-language applications.
The launch comes as enterprises continue exploring voice-enabled AI systems to automate customer interactions and deliver digital services in regional languages.
With its latest release, Sarvam AI is expanding the technical and commercial options available to developers building multilingual voice applications for the Indian market.
Disclaimer: This article may include information derived from interviews, press releases, public statements, research, company communications and other publicly available or third-party sources. Such material may be summarised, paraphrased or contextualised for journalistic and editorial purposes. All rights in third-party content remain with their respective owners.