This Startup Wants to Make Voice Programmable for Developers

Voice is fast becoming a core interface for AI, particularly in multilingual markets like India. As enterprises move from static content generation to real-time, conversational systems, the need for low-latency, scalable voice infrastructure is becoming critical.

Bengaluru-based Murf is positioning itself at the centre of this shift. What began as a voiceover generation tool has evolved into a full-stack voice infrastructure platform designed to power real-time, multilingual AI interactions for enterprises.

The company’s early product, Murf Studio, simplifies voiceover creation, enabling users to convert scripts into human-like audio. The founder says the adoption initially came from content-heavy workflows such as marketing, training, and media production, areas where voice content needs to be produced at scale.

The startup claims it's used by over 300 Forbes Global 2000 enterprises, including Nestlé, Omnicom Production, Pfizer, and Vertiv. Within organisations, teams in learning and development and marketing are increasingly relying on it to create training modules and expand voice content, thereby compressing timelines that historically took much longer.

As enterprise needs shift towards real-time interaction, Murf says it is expanding beyond content generation into infrastructure. It now offers capabilities such as voice experiences, automation, and multilingual dubbing, powered by proprietary speech foundation models.

This transition reflects a broader industry movement, from tools that generate content to systems that enable continuous, interactive AI experiences.

Speed as a Differentiator

At the core of Murf’s infrastructure play is Falcon, its latest model built for real-time speech applications. In conversational AI, latency is critical. Delays beyond a few hundred milliseconds can significantly degrade user experience.

Murf claims a time-to-first-byte of around 130 milliseconds, well below the roughly 500-millisecond threshold required for seamless interaction.

“In conversational AI, responses need to happen within about 500 milliseconds. We reduce the text-to-speech portion to about 130 milliseconds,” Co-founder and COO Sneha Roy says.

The company takes a distinct technical approach. Instead of relying on large, general-purpose models, it builds language-specific architectures, reducing model size by nearly 50%. This improves both speed and cost efficiency.

“We operate at roughly one-sixth the cost of competitors while maintaining healthy margins,” Roy says.

Murf currently supports 35 languages, including 10 Indian languages, positioning it strongly in multilingual and emerging markets.

Business Model and Growth

Murf has raised around $11.5 million (~₹95 crore) from investors, including Z47 and Elevation Capital.

Its revenue model mirrors its product evolution. While its earlier SaaS offerings follow a subscription model, newer infrastructure products like Falcon are priced on a pay-as-you-go basis aligned with API usage.

“Our revenue model has two parts. Customers usually start with one use case and then scale, so flexibility is important,” Roy says.

The company reports a user base of around 10 million creators and developers, with close to one million minutes of audio generated on its platform every month. Roy says its API business is also gaining traction, processing over 110 million characters monthly, translating to roughly 1,30,000 minutes of generated audio.

Adoption varies by use case. While Murf Studio continues to see strong traction in marketing and training, its API offerings are gaining early momentum in sectors such as banking and financial services. Voice agents are being deployed for high-volume outbound use cases such as lead qualification and collections.

Enterprises report measurable gains. Content production cycles that previously took four to six weeks are reduced to hours, with turnaround times improving by 45% to 90%. The founder says some customers reported up to 80% reduction in annual voiceover production costs.

On the API side, Murf-powered voice agents handle upwards of 15,000 outbound calls per day for certain lending use cases, replacing traditional call centre workloads.

From Founders to Infrastructure

Murf was founded in 2020 by IIT Kharagpur batchmates Roy, Ankur Edkie and Divyanshu Pandey. The founding team brings a mix of go-to-market and deep tech experience.

Roy, who leads go-to-market, began her career at ITC before building and exiting an early e-commerce venture in India. She later spent six years at Urban Ladder, where she led marketing and faced challenges with multimedia content creation at scale.

Ankur, her co-founder and spouse, has a research background and previously led AI research at Goldman Sachs.

“We combined our experiences, my exposure to multimedia challenges and his expertise in AI, to build something meaningful in voice,” Roy says.

With a team of approximately 95 employees, around 40% of whom are engineers, and offices in San Francisco and Bengaluru, Murf is working towards its long-term vision.

Challenges and the Road Ahead

Murf operates in a crowded market. On one end are Big Tech players like Amazon, Google, and Microsoft, which bundle voice AI into broader cloud ecosystems. On the other hand, there are specialised startups such as ElevenLabs, OpenAI, Deepgram, and Cartesia, focused on voice generation and real-time speech systems.

Breaking into enterprise infrastructure comes with challenges. The company must navigate long sales cycles, prove reliability at scale, and differentiate itself against well-established players.

It also faces the task of defining its category, moving from a ‘tool’ to an ‘infrastructure layer’ in a market that is still evolving around voice in AI systems.

There are cultural shifts underway as well.

“Often I’m the only woman in the room. But it’s changing slowly,” Roy says.