Sarvam AI Launches Bulbul-v2 with Realistic Indian Accents

Bengaluru-based AI startup Sarvam AI has launched its latest text-to-speech (TTS) model, Bulbul-v2, designed specifically to cater to India’s diverse linguistic landscape. Supporting 11 Indian languages, the model offers natural, human-like voices with authentic regional accents, making it a promising tool for businesses, brands, and developers seeking voice customisation for Indian audiences.

Why in News?

Sarvam AI’s release of Bulbul-v2 marks a significant leap in India’s speech AI ecosystem, offering India-first pricing, low latency, and custom voice options. The model’s launch also aligns with the company’s role in developing India’s sovereign large language model under the IndiaAI mission.

Key Features of Bulbul-v2

• Supports 11 Indian languages with regional accent precision.
• Enables real-time synthesis and multi-language (including code-mixed) text support.

Fine-grained control over,

• Pitch
• Pace
• Loudness
• Multiple sample rates (8kHz to 24kHz).
• Smart text preprocessing: normalises numbers, dates, and mixed-language content.

Aim and Objectives

• To democratise AI voice technology for Indian users.
• Offer a customisable, natural-sounding voice model suitable for various business and branding applications.
• Promote linguistic inclusivity in India’s digital ecosystem.

Background

• Bulbul-v1 was launched in August 2024 with six preset voice personalities.
• Sarvam AI became the first Indian startup selected to develop a sovereign Indian LLM with reasoning and voice capabilities.

Significance

• Improves accessibility for digital services in local languages.
• Enables brands to reach regional audiences with authentic-sounding voices.
• Boosts India’s AI ecosystem, supporting the IndiaAI mission for technological self-reliance.