Ukrainian start-up Respeecher secures $1M to develop AI voice cloning responsibly

Respeecher is a start-up that came into the public eye after Disney tasked it with recreating the voice of James Earl Jones for Darth Vader after the actor stepped down from the role. In a display of perseverance in the face of adversity, the Respeecher team scrambled to get the job done just as Russia was invading Ukraine on February 24, 2022. Respeecher had already worked with Lucasfilm in developing the voice of a younger Luke Skywalker for Disney+ The Book of Boba Fett and counts game developer CD Projekt and Warner Music among its clients.

Respeecher's work on recreating Darth Vader's voice exemplifies the importance the company places on its ethical principles. In line with these principles, Lucasfilm sound editor Matthew Wood showed Jones the work Respeecher had done in the past before moving forward. Once he was familiar with the company, Jones signed the consent so the company could access archival recordings of his voice so Darth Vader could be kept alive by any means necessary.

Other aspects of its ethical principles entail that Respeecher does not provide a public API for voice creation, the company only works with trusted clients, and it is working on watermarking technologies that make it possible to identify whether a piece of audio was generated using Respeecher technology or not. The company has also set new goals for itself concerning educating the general public on the capabilities of synthetic speech, developing technologies to recognize any AI-generated audio content, and working with companies such as Google and Meta to mitigate the risk posed by AI-powered voice generation on social media.

In addition to its ethical principles, Respeecher's unique methodology for synthetic voice creation also sets it apart from its competitors. Rather than depending on text-to-speech generators, the Respeecher depends on a speech-to-speech transformation that converts a human-sourced voice into the target voice. Once clients are granted permission to use the target voice (by the living people, their families, or their estate), they must provide high-quality recordings of this voice to Respeecher, optionally in conjunction with an audio source of comparable quality. Respeecher then trains its AI to create voice-swapping models that convert any voice sample into the target voice. Clients can send in their files for in-house conversion, use the private API, or the Respeecher web app.

The results are synthetic voices that sound natural, almost indistinguishable from the original. Respeecher synthetic voices lack the robotic, emotion-lacking quality that is often associated with AI-powered voice generation. Moreover, unlike traditional text-to-speech approaches, speech-to-speech generated voices can easily be modified to account for or eliminate accents, foreign words, or the typical nuances found in human speech, such as humming or giggling.

The pre-Series A funding round that brought in $1M counted with participation from entrepreneur Gary Vaynerchuk and funds ffVC, Bad Ideas, ICU, and SID Venture Partners.