Wispr's $280M bet: voice input's problem is accuracy, not appetite

Wispr said Monday it raised $280 million in Series B funding at a $2 billion valuation, led by Menlo Ventures, taking total funding since its 2021 founding to $361 million ^[1][2]. The money goes to two things at once: a first in-house speech model called Canto, and a move out of dictation into meeting notes with a product called Notetaker ^[3][4].

The size is the tell. Wispr had raised roughly $81 million before this round and closed its previous one less than ten months ago ^[5], so the Series B is about 3.5 times everything that came before it ^[6]. The company says people have written more than 60 billion words with Flow, and that almost all Fortune 500 companies and more than 10,000 enterprises use the app ^[7]. Whatever else is unproven here, appetite for talking to a text field is not the open question. The open question is the transcript. "One wrong word sends users back to the keyboard and breaks their focus," CEO and co-founder Tanay Kothari wrote in a company blog post ^[8].

Canto is the answer to that, and the numbers deserve reading twice. Wispr says word error rates in the hardest conditions, meaning background noise, wind, heavy accents, or music, fall from more than 30 percent to between 5 and 10 percent, which it calls a more than 4x reduction ^[9]. At the top of that new range, a 20-word dictation still averages two wrong words ^[11]. The company also expects 30 to 35 percent fewer dictations to need editing in daily use ^[10], which leaves 65 to 70 percent of them still needing it ^[12]. Worth noting that Canto was previewed after several users complained that Flow's dictation had gotten worse ^[13], so part of the gain is repair. Accuracy is also the axis on which commodity models improve for free, and Wispr already faces Willow, Monologue, Aqua and Superwhisper plus a wave of cheaper and free tools aimed at power users ^[14].

The meeting-notes move is the riskier half. Notetaker shipped about a week before the raise, records calls and in-person conversations, labels speakers on a live transcript, and produces topic-organized summaries with dates, decisions and next steps ^[4]. That puts Wispr against Granola, Fireflies, Read AI and Otter ^[15]. Two of those four, Otter and Granola, are facing separate California lawsuits alleging privacy violations ^[16][17]. Wispr's architectural choice is pointed: CTO and co-founder Sahaj Garg says Notetaker records audio locally and does not join calls as a visible bot, which he acknowledges puts the burden on users to disclose in states where consent is required ^[18]. That is a cheaper legal posture for Wispr and a more expensive one for the enterprise buyers it says it already has.

Around this, Wispr set up an Advanced Interfaces Lab under Chief Scientist Ariya Rastrow, a founding member of the team that built Amazon's Alexa, to work on interfaces that respond to what a user says rather than only transcribing it ^[19]. Existing backers Notable Capital, NEA, Neo Ventures, 8VC and MVP Ventures joined, alongside new investors Acrew, Forerunner, Goodwater, Peak XV, Together Fund and PLUS Capital ^[20], plus athletes including Livvy Dunne, Shaun White, Dak Prescott, Joe Burrow, Klay Thompson, Paul George and Trae Young ^[21]. Domantas Sabonis, an investor, said Flow keeps up as he switches between English, Spanish and Lithuanian ^[22].

Watch whether Canto's error-rate claims hold outside the company blog, whether enterprise legal teams accept a design that shifts recording disclosure onto employees, and what the Otter and Granola cases establish about tools that never announce themselves on the call.