Jump to content

15.ai

From Simple English Wikipedia, the free encyclopedia
15.ai
Type of site
Artificial intelligence, speech synthesis, generative artificial intelligence
Available inEnglish
Owner15
Founder(s)15
URL15.ai (redirects to 15.dev)
CommercialNo
RegistrationNone
LaunchedMarch 2020; 6 years ago (2020-03)
Current statusActive

15.ai is a website that uses artificial intelligence (AI) to make fictional characters speak. It is free for anyone to use. 15.ai was created by an AI researcher who uses the alias 15. They started making it while they were studying at the Massachusetts Institute of Technology (MIT). 15.ai lets people type in any text. It makes popular characters speak it with different emotions. 15.ai can copy voices using only a small amount of speech. The name "15.ai" comes from the creator saying that a voice can be copied with just 15 seconds of audio. It was one of the first websites to use generative artificial intelligence.

15.ai started in March 2020. It became very popular in early 2021. Many Internet fan groups used it. Fans of My Little Pony: Friendship Is Magic, Team Fortress 2, Portal, and SpongeBob SquarePants used it often. 15.ai let users control emotions using emojis. It also let users control how words are pronounced. People who reviewed the website liked it. They said it was easy to use. They also liked how well it controlled emotions. However, people said it sometimes had problems with speech rhythm. Some people also said they did not like how it only worked in English. 15.ai is the first website to make AI voices popular for Internet memes and making online content.

Voice actors talked about what 15.ai meant for their work. They said it could take away jobs. They also said it could be used to trick people and steal from them. In January 2022, a company called Voiceverse made voices using 15.ai. Voiceverse sold the voices as non-fungible tokens (NFTs). Voiceverse did not tell people they used 15.ai. They said their own technology made the voices. Later, people discovered the truth. News websites said the company stole from 15.ai.

15.ai stopped in September 2022 because of legal problems. After it shut down, voice AI companies made similar websites that needed money to use. These companies said that 15.ai was important in voice AI history. On May 18, 2025, the creator launched 15.dev as the new version of 15.ai.

Background

[change | change source]

Speech synthesis means making computers talk like humans. For many years, computers did not sound natural when they talked. The old way to make computer speech was called concatenative synthesis. This method worked by cutting up recordings of real human voices into small pieces. The computer would then put these pieces together to make new sentences. This often made the speech sound robotic.[1]

In 2016, a company called DeepMind made a new system called WaveNet. WaveNet used a type of AI called a neural network to make speech. Neural networks are computer programs that learn patterns from data. WaveNet made much better speech than the old method. In 2018, Google made an AI system called Tacotron 2. Tacotron 2 could make very natural-sounding speech. However, it needed a lot of data to work well. It needed about 24 hours of audio recordings to make good speech. With only 24 minutes of audio, it could not make speech that people could understand.[2] Other systems also improved speech synthesis around this time, such as HiFi-GAN[3] and Glow-TTS.[4] However, all of these systems needed many hours of audio recordings. This made it hard for regular people to use the technology.[1]

2016–2020: Conception and development

[change | change source]

The creator of 15.ai uses the name 15 as an alias. 15 thought of the idea for 15.ai in 2016. It started as a research project in AI speech synthesis. At that time, 15 was a freshman (first-year student) at MIT. 15 was inspired by DeepMind's WaveNet system. They kept working on the project while they studied at MIT. By 2019, 15 showed that their system needed 75% less training data than WaveNet and Tacotron 2. They said a voice can be copied with only 15 seconds of speech. This is why it was called "15.ai".[6]

15 originally planned to get a PhD (an advanced degree). The PhD would be based on their research. However, 15 chose to work at a technology company instead. This happened after their company got into Y Combinator in 2019. Y Combinator is a program that helps new technology companies grow. 15 left the company in early 2020. After they left, 15 returned to their research. They started making it into a web application (a program that runs in a web browser). To test it, 15 wanted to use voices with a lot of emotions. This way, it would show that the AI could handle difficult emotions and complex speech.[6] Soon, 15 found the Pony Preservation Project (PPP). The PPP was a project started by /mlp/, the My Little Pony board on 4chan. People who did not use their real names cut out every single voice line from My Little Pony: Friendship Is Magic. This included all nine seasons of the show and the movie. Then they wrote down what each character said, which character said it, and what emotion they had while saying it. They put all of this data into a dataset (an organized collection of data). This dataset was important because it was what 15 needed to test their AI.[7]

2020–2022: Release and operation

[change | change source]

15.ai started on March 2, 2020.[8] Users do not have to register or pay money to use it. It only makes users accept a terms of service (a set of rules) before using it.[9] In the beginning, 15.ai only had a few characters, such as Twilight Sparkle from Friendship Is Magic and the Soldier from Team Fortress 2.[10] Users are allowed to use the character voices for anything they wanted. They only had to give credit by writing "15.ai" somewhere if they posted it online.[11] They were also not allowed to mix 15.ai's voices with other AI-generated voices. This was to stop confusion over which AI made the voices.[12] In late 2020, 15 added a multi-speaker embedding to the neural network. This is a way for the AI to train on many different voices at the same time.[6] Because of this, 15 added more than 50 characters to 15.ai in a few months.[8] This also let the AI learn emotions for a character even when those emotions were not in the training data.[11]

By May 2020, 15.ai had made over 4.2 million voice lines.[13] In early 2021, 15.ai became very popular.[14] Viral videos made using 15.ai's voices spread on Twitter, TikTok, Reddit, Twitch, Facebook, and YouTube.[15] These videos included skits, memes, and fan content.[14] 15.ai sometimes cost US$12,000 a month to run.[16] This was because millions of voice lines were being made in a single day.[6] Some companies offered to buy 15.ai. However, 15 kept 15.ai free by paying for it with their own money. 15 was able to do this because of the money they earned from the company they started in 2019.[6]

2022: Voiceverse NFT scandal

[change | change source]
A meme about how NFTs can be "right-clicked and saved". People said that it was ironic that Voiceverse wanted to sell AI voices. This was because Voiceverse had stolen them from 15.ai.

In December 2021, 15 said that they did not want to use NFTs with 15.ai.[17] A month later, on January 14, 2022, a famous voice actor named Troy Baker said he would work with a voice AI company. The company, called Voiceverse, posted videos that had AI-generated voices.[18] Voiceverse said their own technology made these voices.[18] They sold these voices as NFTs.[19]

On the same day, 15 said that Voiceverse sounded like a scam.[20] A few hours later, 15 said that Voiceverse was lying. 15 looked at 15.ai's log files (records of who used the website and what they did). They found that someone had used 15.ai to make the same voice lines that were in Voiceverse's videos.[21] This broke 15.ai's terms of service. This was because Voiceverse used the voices to make money without giving credit to 15.ai.[9] 15 posted these log files to Twitter as proof.[22]

Because of this, Voiceverse admitted that the voices were made by 15.ai. Voiceverse blamed the people on their team who make advertisements.[23] Voiceverse said they used 15.ai because their own technology was not good enough.[24] 15 responded with "Go fuck yourself."[25] Many people supported 15's response. It got hundreds of thousands of retweets and likes on Twitter.[6] Voiceverse then deleted the videos that had the stolen voices.[26]

Troy Baker Twitter
@TroyBakerVA

I'm partnering with @VoiceverseNFT to explore ways where together we might bring new tools to new creators to make new things, and allow everyone a chance to own & invest in the IP's they create.
We all have a story to tell. You can hate. Or you can create.
What'll it be?

January 14, 2022[27]

Aftermath

[change | change source]

This incident made people worry about NFT projects. People said that NFT projects were often connected to stealing and scamming.[28] Many news websites said that what Voiceverse did was theft.[a] The AI Incident Database (AIID) wrote that Voiceverse stole from 15.ai.[49] The AI, Algorithmic, and Automation Incidents and Controversies (AIAAIC) also wrote about Voiceverse copying 15.ai's work without permission.[50] Molly White wrote about it on her website called Web3 Is Going Just Great.[32]

Pavel Khibchenko of Skillbox wrote that this was an example of fraud in NFTs.[51] Voice actor and YouTuber Yong Yea said that Voiceverse took 15.ai's better technology on purpose. Yong Yea said Voiceverse did this to make their own voice NFTs look better than they really were.[38] Troy Baker was also criticized for supporting the NFT project. People did not like the way he announced it. Baker later said that his original tweet was "antagonistic" (meaning it was rude and tried to start a fight).[52] On January 31, 2022, Baker said he would stop working with Voiceverse.[53]

In 2024, a class action lawsuit (a court case where many people sue together) was filed against LOVO, Inc., which is the parent company (owner) of Voiceverse. The court papers mentioned that Voiceverse had stolen 15.ai's technology as part of the case.[54]

2022–2024: Inactivity

[change | change source]

In September 2022, 15.ai was taken offline.[55] Even though 15.ai was offline, voice AI companies continued to say that 15.ai was important. PlayHT, a Y Combinator startup, called 15.ai's first release "a breakthrough" in text-to-speech technology.[56] Cliff Weitzman, who started a company called Speechify, said that 15.ai made AI voice copying popular. He said it was the first website to let people use voices of famous fictional characters.[57] Mati Staniszewski, who started a company called ElevenLabs, wrote that 15.ai changed the field of AI text-to-speech.[58]

2025–present: Revival

[change | change source]

On May 18, 2025, 15 launched 15.dev as the official follow-up to 15.ai.[59] At first, 15.dev only had characters from My Little Pony: Friendship Is Magic. 15 said on Twitter that more characters and features would be added later.[tweet 1] Fan news website Equestria Daily wrote that the new website had voices for almost every pony character from the show. It also had a menu where users could pick different emotions for the voices.[59]

[change | change source]
  1. The following news publications said this: The Mary Sue,[29] Kotaku Australia,[9] Eurogamer,[19] NME,[26] Muropaketti,[30] Gamereactor,[31] Web3 Is Going Just Great,[32] StopGame,[33] iXBT Games,[34] DTF,[35] Sport.es,[36] FZ,[37] The Journal,[38] LevelUp,[39] Stevivor,[40] PlayStation Universe,[41] Checkpoint Gaming,[20] Tech Times,[42] Mobidictum,[43] OtakuPT,[44] Gamebrott,[45] GameGuru,[46] Shazoo,[47] and Geek Culture.[48]

References

[change | change source]
  1. 1 2 Barakat, Turk & Demiroglu 2024.
  2. Google 2018
  3. Kong, Kim & Bae 2020.
  4. Kim et al. 2020.
  5. "Examples". May 15, 2025. Retrieved May 16, 2025.[permanent dead link]
  6. 1 2 3 4 5 6 Temitope 2024.
  7. Branwen 2020.
  8. 1 2 Abisola 2025.
  9. 1 2 3 Innes 2022.
  10. Ng 2020.
  11. 1 2 Kurosawa 2021.
  12. "About". 15.ai (Official website). March 2, 2020. Retrieved December 23, 2024.{{cite web}}: CS1 maint: deprecated archival service (link)
  13. Mahendra 2020.
  14. 1 2 Zwiezen 2021; Ruppert 2021; Clayton 2021; Morton 2021.
  15. Ruppert 2021; Abisola 2025.
  16. Temitope 2024; Abisola 2025.
  17. Lopez 2022; Innes 2022; Piletsky 2022.
  18. 1 2 Phillips 2022; Williams 2022.
  19. 1 2 Phillips 2022.
  20. 1 2 Wakelin-King 2022.
  21. Phillips 2022; Innes 2022; Wright 2022.
  22. Phillips 2022; Innes 2022.
  23. Phillips 2022; Myrén 2022; AIAAIC 2022.
  24. Kamen 2022.
  25. Wright 2022; Groth-Anderson 2022; Myrén 2022; Archer 2022; Williams 2022.
  26. 1 2 Williams 2022.
  27. Innes 2022; Enriquez 2022.
  28. Коэн 2022; Baylos 2022.
  29. Lawrence 2022.
  30. Muropaketti 2022.
  31. Groth-Anderson 2022.
  32. 1 2 White 2022.
  33. Skorich 2022.
  34. Piletsky 2022.
  35. Granger 2022.
  36. Baylos 2022.
  37. Myrén 2022.
  38. 1 2 Parker 2022.
  39. Rosas 2022.
  40. Wright 2022.
  41. Carcasole 2022.
  42. Henry 2022.
  43. Aktaş 2022.
  44. Archer 2022.
  45. Ifram 2022.
  46. Kuchkanov 2022.
  47. Коэн 2022.
  48. Toh 2022.
  49. Sonali 2022.
  50. AIAAIC 2022.
  51. Khibchenko 2022.
  52. Wright 2022; White 2022; Carcasole 2022; Enriquez 2022.
  53. Kamen 2022; Granger 2022.
  54. Paul Lehrman and Linnea Sage, et al. v. LOVO, Inc., No. 1:24-cv-03770, 38 (S.D.N.Y. 2024) (“Separately, VoiceVerse has already been found to have stolen technology from another company. See Ule Lopez, WCCF Tech, "Voiceverse NFT Service Reportedly Uses Stolen Technology from 15ai," (Jan. 16, 2022), https://wccftech.com/voiceverse-nft-service-uses-stolen-technology-from-15ai/.”).
  55. Staniszewski 2024; Temitope 2024.
  56. Play.ht 2024.
  57. Weitzman 2023.
  58. Staniszewski 2024.
  59. 1 2 Scotellaro 2025.
  1. @fifteenai (May 18, 2025). "We are so back. 15.dev Only MLP characters for now. More characters, features, and improvements will be added soon" (Tweet). Retrieved May 18, 2025 via Twitter.{{cite web}}: CS1 maint: deprecated archival service (link)

Works cited

[change | change source]

Other websites

[change | change source]