Again in 2016, Hammad Syed and Mahmoud Felfel, an ex-WhatsApp engineer, thought it’d be neat to construct a text-to-speech Chrome extension for Medium articles. The extension, which might learn any Medium story aloud, was featured on Product Hunt. A yr later, it spawned a whole enterprise.
“We noticed an even bigger alternative in serving to people and organizations create lifelike audio content material for his or her functions,” Syed advised TechCrunch. “With out the necessity to construct their very own mannequin, they might deploy human-quality speech experiences sooner than ever earlier than.”
Syed and Felfel’s firm, PlayAI (previously PlayHT), pitches itself because the “voice interface of AI.” Prospects can select from numerous predefined voices, or clone a voice, and use PlayAI’s API to combine text-to-speech into their apps.
Toggles permit customers to regulate the intonation, cadence, and tenor of voices.
PlayAI additionally provides a “playground” the place customers can add a file to generate a read-aloud model and a dashboard for creating more-polished audio narrations and voiceovers. Lately, the corporate bought into the “AI brokers” recreation with instruments that can be utilized to automate duties comparable to answering buyer calls at a enterprise.
Certainly one of PlayAI’s extra fascinating experiments is PlayNote, which transforms PDFs, movies, photographs, songs, and different recordsdata into podcast-style reveals, read-aloud summaries, one-on-one debates, and even youngsters’s tales. Like Google’s NotebookLM, PlayNote generates a script from an uploaded file or URL and feeds it to a group of AI fashions, which collectively craft the completed product.
I gave it a whirl, and the outcomes weren’t half unhealthy. PlayNote’s “podcast” setting produces clips roughly on par with NotebookLM’s when it comes to high quality, and the device’s potential to ingest photographs and movies makes for some fascinating creations. Given an image of hen mole dish I had not too long ago, PlayNote wrote a five-minute podcast script about it. Really, we live sooner or later.
Granted, the device, like all AI instruments, generates odd artifacts and hallucinations once in a while. And whereas PlayNote will do its greatest to adapt a file to the format you’ve chosen, don’t count on, say, a dry authorized submitting to make for the perfect supply materials. See: the Musk v. OpenAI lawsuit framed as a bedtime story:
PlayNote’s podcast format is made doable by PlayAI’s newest mannequin, PlayDialog, which Syed says can use the “context and historical past” of a dialog to generate speech that displays the dialog move. “Utilizing a dialog’s historic context to regulate prosody, emotion, and pacing, PlayDialog delivers dialog with pure supply and acceptable tone,” he continued.
PlayAI, which is shut rivals with ElevenLabs, has been criticized previously for its laissez faire method to security. The corporate’s voice cloning device requires that customers examine a field indicating that they “have all the required rights or consent” to clone a voice — however there isn’t any enforcement mechanism. I had no hassle making a clone of Kamala Harris’ voice from a recording.
That’s regarding contemplating the potential for scams and deepfakes.
PlayAI additionally claims that it routinely detects and blocks “sexual, offensive, racist, or threatening content material.” However that wasn’t the case in my testing. I used the Harris clone to generate speech I frankly can’t embed right here and by no means as soon as noticed a warning message.
In the meantime, PlayNote’s group portal, which is crammed with publicly generated content material, has recordsdata with express titles like “Lady Performing Oral Intercourse.”
Syed tells me that PlayAI responds to stories of voices cloned with out consent, like this one, by blocking the person accountable and eradicating the cloned voice instantly. He additionally makes the case that PlayAI’s highest-fidelity voice clones, which require 20 minutes of voice samples, are priced larger ($49 per 30 days billed yearly or $99 per 30 days) than most scammers are prepared to pay.
“PlayAI has a number of moral safeguards in place,” Syed mentioned. “We’ve applied sturdy mechanisms to determine whether or not a voice was synthesized utilizing our expertise, for instance. If any misuse is reported, we promptly confirm the origin of the content material and take decisive actions to rectify the scenario and stop additional moral violations.”
I’d definitely hope that’s the case — and that PlayAI strikes away from advertising and marketing campaigns that includes useless tech celebrities. If PlayAI’s moderation isn’t sturdy, it might face authorized challenges in Tennessee, which has a legislation on the books stopping platforms from internet hosting AI to make unauthorized recordings of an individual’s voice.
PlayAI’s method to coaching its voice-cloning AI can be a bit murky. The corporate gained’t reveal the place it sourced the information for its fashions, ostensibly for aggressive causes.
“PlayAI makes use of principally open knowledge units, [as well as licensed data] and proprietary knowledge units which are constructed in-house,” Syed mentioned. “We don’t use person knowledge from the merchandise in coaching, or creators to coach fashions. Our fashions are educated on tens of millions of hours of real-life human speech, delivering voices in female and male genders throughout a number of languages and accents.”
Most AI fashions are educated on public net knowledge — a few of which can be copyrighted or beneath a restrictive license. Many AI distributors argue that the fair-use doctrine shields them from copyright claims. However that hasn’t stopped knowledge homeowners from submitting class motion lawsuits alleging that distributors used their knowledge sans permission.
PlayAI hasn’t been sued. Nevertheless, its phrases of service counsel it gained’t go to bat for customers in the event that they discover themselves beneath authorized risk.
Voice cloning platforms like PlayAI face criticism from actors who concern that voice work will finally get replaced by AI-generated vocals, and that actors could have little management over how their digital doubles are used.
The Hollywood actors’ union SAG-AFTRA has struck offers with some startups, together with on-line expertise market Narrativ and Reproduction Studios, for what it describes as “truthful” and “moral” voice cloning preparations. However even these tie-ups have come beneath intense scrutiny, together with from SAG-AFTRA’s personal members.
In California, legal guidelines require corporations counting on a performer’s digital reproduction (e.g. cloned voice) give an outline of the reproduction’s supposed use and negotiate with the performer’s authorized counsel. In addition they require that leisure employers achieve the consent of a deceased performer’s property earlier than utilizing a digital clone of that particular person.
Syed says that PlayAI “ensures” that each voice clone generated by way of its platform is unique to the creator. “This exclusivity is significant for shielding the inventive rights of customers,” he added.
The growing authorized burden is one headwind for PlayAI. One other is the competitors. Papercup, Deepdub, Acapela, Respeecher, and Voice.ai, in addition to massive tech incumbents Amazon, Microsoft, and Google, provide AI dubbing and voice cloning instruments. The aforementioned ElevenLabs, one of many highest-profile voice cloning distributors, is alleged to be elevating new funds at a valuation over $3 billion.
PlayAI isn’t struggling to search out buyers, although. This month, the Y Combinator-backed firm closed a $20 million seed spherical led by Kindred Ventures, bringing its complete capital raised to $21 million. Race Capital and 500 World additionally participated.
“The brand new capital might be used to spend money on our generative AI voice fashions and voice agent platform, and to shorten the time for companies to construct human-quality speech experiences,” Syed mentioned, including that PlayAI plans to broaden its 40-person workforce.
