Lyrebird

Overview

Lyrebird was a Montreal-based AI startup that built voice-cloning technology capable of replicating any person's voice from just one minute of audio.Founded in 2017 by three PhD students at MILA—one of the world's leading deep learning labs—the company commercialized its own academic research and generated enormous public attention with a viral launch demo.

Despite genuine technical achievement and early investor interest from top-tier firms, Lyrebird raised only $120K in total disclosed funding, never publicly launched the developer API that was its intended revenue engine, and was acquired by Descript in September 2019 for an undisclosed sum.The core thesis of failure: Lyrebird built a real technology with no product context to deploy it in.

Voice cloning as a standalone API had no clear killer use case in 2017–2019; voice cloning embedded inside a content creation workflow did.Descript provided that context, the capital to execute, and the distribution to reach paying users—none of which Lyrebird could assemble independently.

Image 1 / 2

Founding Story

Lyrebird emerged directly from the research labs of MILA, the Montreal Institute for Learning Algorithms, one of the world's premier deep learning research centers. The three co-founders—Alexandre de Brébisson, Kundan Kumar, and Jose Sotelo—were PhD students in artificial intelligence at the University of Montreal, working under the supervision of Yoshua Bengio, Pascal Vincent, and Aaron Courville, three of the most influential figures in modern deep learning.^{[1] [2]}

The founding team did not set out to build a startup in the conventional sense. Their voice synthesis research was academic in origin—part of the broader wave of neural network breakthroughs happening at MILA in the mid-2010s that would eventually reshape the field of speech synthesis. The decision to commercialize came from recognizing that the technology had crossed a threshold: one minute of audio was now sufficient to generate a convincing digital replica of a human voice, with emotional modulation and near-real-time generation speed.^[3]

The team bootstrapped development, continuing to work within the MILA lab infrastructure, and made a deliberate choice to ship a public demo before seeking external capital.^[4] This sequencing—demonstrate first, raise second—reflected both scrappiness and a research-lab orientation: prove the technology works, then figure out the business. Alexandre de Brébisson served as CEO, with Kundan Kumar and Jose Sotelo rounding out the technical founding team.^[5]

The team was accepted into Y Combinator's Summer 2017 batch, receiving the standard $120K seed check.^[6] YC provided validation and a network, but the funding was thin for a company attempting to build and maintain a real-time voice synthesis API platform at commercial scale.

The founders described themselves in their YC launch post as "co-founders of Lyrebird and PhD students in AI at University of Montreal" building "speech synthesis technologies to improve the way we communicate with computers."^[7] The framing was deliberately broad—a platform vision rather than a specific product. That breadth would prove both a strength and a liability.

De Brébisson articulated the long-term vision with a prescient analogy: "The situation is comparable to Photoshop. People are now aware that photos can be faked. I think in the future, audio recordings are going to become less and less reliable [as evidence]."^[8] He was right about the trajectory. Being right about the future, however, did not solve the near-term problem of building a sustainable business in 2017.