ElevenLabs’ new v4 speech model supports more expression control and 90 languages

ElevenLabs launched 2 caller code models connected Monday, called ElevenLabs v4 and v4 Turbo, offering much look control, little latency for dependable agents, and enactment for much than 90 languages.

The institution released its v3 exemplary past year, and teased the exemplary astatine an lawsuit successful Warsaw earlier this year. For the v4 procreation of models, ElevenLabs is adopting a caller architecture that allows for amended power and faster cloning. The institution said that with v4, users volition beryllium capable to clone a dependable with conscionable 10 seconds of audio.

On the originative side, the exemplary handles dependable individuality amended implicit longer chunks of text. It besides keeps the discourse of the substance successful caput portion speechmaking it aloud to alteration expressions. ElevenLabs introduced inline tags to specify look with v3, and is expanding those tags successful v4, letting users stack aggregate tags and having the exemplary travel the sequence.

The erstwhile mentation supported 70 languages, and ElevenLabs has worked to get that fig up to 90 languages with the caller version. The startup said that it observed the biggest prime leap successful Japanese, Brazilian Portuguese, Mandarin and Cantonese.

ElevenLabs has scaled its endeavor calling concern rapidly implicit the past year, with much than 55% of its concern coming from ample companies. The institution said that the caller exemplary is suited for dependable agents, arsenic the caller mentation has little latency to let for much fluid conversation. Plus, the v4 tin commencement generating audio arsenic soon arsenic the LLM down it starts generating answers. What’s more, the exemplary tin besides grip confrontations, escalations, and holds otherwise for amended contented resolution.

Competition successful code models has ramped up arsenic startups similar Cartesia, Deepgram, Fish Audio, Boson, and WellSaid Labs person created expressive code models. Big companies similar Google and OpenAI person besides improved their dependable models.

ElevenLabs raised $500 cardinal earlier this twelvemonth from Sequoia earlier this year, successful a circular led by Sequoia that valued the institution astatine $11 billion. There are already rumors of a followup fundraising circular that would worth the institution astatine $22 billion. ElevenLabs annualized gross tally complaint has climbed from astir $330 cardinal astatine the commencement of the twelvemonth to implicit $600 million. The institution has aggressively hired unit successful antithetic markets specified arsenic India, Europe, and Brazil, arsenic its headcount has reached implicit 800.

In a caller interrogation with TechCrunch, the company’s co-founder and CEO Mati Staniszewski said that the institution is aiming for an IPO “in the adjacent years,” but didn’t perpetrate to a timeline.