Nat弗裏德曼

2023年6月20日

啟動使用生成AI創建合成的聲音為我們從電影配音AI同伴應用4分鍾讀美國的一些主要投資者推倒門今年歐洲衍生的AI初創公司。今年5月,NEA主動提供了斯德哥爾摩的薩那實驗室,上周支持倫敦Synthesia Nvidia公司“沒有積極尋找新的投資”。本周,它的總部設在紐約,倫敦和ElevenLabs——創業建築技術來生成合成聲音——引發了一係列19美元由Andreessen Horowitz, GitHub Nat創始人弗裏德曼和丹尼爾•格羅斯,前蘋果的機器學習總監。ElevenLabs創始人馬蒂斯著名Staniszewski也告訴篩過,他不是“積極”的投資,但與總值早期對話之後,弗裏德曼和a16z,意識到這些重量級的支持者能夠幫助業務規模。廣告構建B2B提供Staniszewski說,錢將用於雇傭人在兩個領域:機器學習專家可以幫助改善ElevenLabs核心音頻技術,軟件工程師和產品設計師誰可以建立跨三個垂直的公司提供的。他說,這些出版(有聲讀物,時事通訊和voiced-up篩選文章,我們已經嚐試了),遊戲(遊戲角色AI-generated聲音)和娛樂(電影和電視AI配音)。Staniszewski表示,自產品推出了今年早些時候,最快的收縮已經從個別出版商,像youtube用戶創建配音或獨立的書的作者來說,得到一個voice-acted有聲讀物記錄會十分昂貴。他說,根據書的長度,它通常成本不超過幾百美元”生成一個有聲讀物和一個人工智能的聲音。Staniszewski說ElevenLabs現在專注於拓展其B2B業務,已經與有聲讀物簽署了合作平台Storytel和數字媒體出版商TheSoul,都加入了一輪作為戰略投資者。“當我們推出的重點都是創造者——人們創造畫外音YouTube或社交媒體,”他解釋說。 “In recent months it's shifted a lot to B2B.” Sifted Newsletters Join to Sign Up Staniszewski adds that he’ll soon be able to announce a partnership with a leading news organisation, which he says will be using the technology to automatically turn its articles into audio. The original vision While publishing appears to be ElevenLabs’ earliest sign of product-market fit, the startup is still working hard to realise its original vision for the company — a comprehensive AI film dubbing system. Staniszewski and his cofounder Piotr Dabkowski are both Polish and started the company to try and build a solution for the low-quality film dubbing that was available in their native language. There are still some challenges to overcome, and not just technical ones. Staniszewski explains that ElevenLabs’ models will have to account for the fact that certain languages tend to use more time and words to communicate the same message. “When you translate something from English to Spanish, it's gonna be about 30-40% longer on average in Spanish,” he says. “You need to think not only about directly translating, but also paraphrasing to make sure that the length is the same.” Advertisement As well as movie dubbing, ElevenLabs is also working on a new — slightly dystopian — application for synthetic voices: AI companion apps that turn chatbot responses into lifelike voices. “We've seen that already in Asian countries like Korea or Japan where that's a popular thing,” he says. To do this, the startup is working on bringing down the latency time of responses from the AI voice, which is currently eight tenths of a second, to half of a second, to make interactions feel more natural. Synthetic voice technology like this also opens the doors to what Staniszewski describes as “nefarious use cases” — think deepfakes or phone scams. He says the company is continually working on safeguards to try and minimise misuse of its platform, and recently released a tool where people can check if audio is AI-generated. What’s the market like? ElevenLabs is one of a number of startups using generative AI to make synthetic voice software. Others include Valencian Voicemod, which is focused on the gaming sector, and London-founded Sonantic, which was acquired by Spotify last year. Big tech is also getting in on the action — Meta recently released its own voice generation tool — so startups will need to hustle hard to make sure they’re responsive to customer needs if they’re to stay in the race. But now, with the backing of some of Silicon Valley’s most reputed investors, ElevenLabs has the level of capital and the guidance to make a real stab at it. Advertisement