Voice morphing seem to be finally catching up, starting from the ones that converted your voice into animal sounds the latest one promises to synthesise your voice completely. Meet Lyrebird, a startup that is launching a new API which will let developers recreate anyone’s voice by scraping an one minute sample of the same. Well, this also reminds me of the Adobe Project VoVO technology which uses system resources to create digital voices while the Lyrebird API uses the cloud resources.
That said Adobe VoCo needs a minimum of a 20-minutes sample to synthesise the voice while the Lyrebird just needs a minute. Lyrebird says that “Our GPU clusters generate 1000 sentences in less than half a second,” claims the company. Furthermore, Lyrebird explains how their API compresses voice DNA into a unique key and this key can be used to generate anything. Interestingly the company also claims that they can control the emotion of the generated voice and also introduce anger, stress, sympathy and other characteristics.
However, the demo sample still feels unfinished and that is because the sample is indeed a work in progress. The demo is a glimpse at the potent of the technology and needless to say the final API is expected to be better than this. Well, a voice synthesiser this advanced is expected to create a privacy mayhem and Lyrebird seems to be well aware of the perils.
By releasing our technology publicly and making it available to anyone, we want to ensure that there will be no such risks. We hope that everyone will soon be aware that such technology exists and that copying the voice of someone else is possible. More generally, we want to raise attention about the lack of evidence that audio recordings may represent in the near future.
Lyrebird further assures that the voice synthesiser will be used for purposes like personal assistants, audio books readers and to synthesise voices of the famous people. Lyrebird’s Artificial Intelligence and the machine learning capabilities are still being put together. Also, Lyrebird has not revealed when it will make its API generally available.