PlayHT
Last reviewed
Jun 4, 2026
Sources
21 citations
Review status
Source-backed
Revision
v1 · 1,936 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 4, 2026
Sources
21 citations
Review status
Source-backed
Revision
v1 · 1,936 words
Add missing citations, update stale details, or suggest a clearer explanation.
PlayHT, later rebranded PlayAI (and reachable at play.ht and play.ai), was an American generative AI voice company that built text-to-speech models, voice cloning tools, and a platform for conversational AI voice agents. Founded by Mahmoud Felfel and Hammad Syed, the company began in 2016 as a browser extension for listening to articles and grew into a developer-focused voice AI platform whose models (Play 2.0, Play 3.0, Play 3.0 mini, and PlayDialog) were used to generate synthetic speech in well over 100 languages. In November 2024 PlayAI raised a $21 million seed round, and in July 2025 Meta acquired the company and brought its roughly 35-person team into its newly formed superintelligence organization. Following the deal, the PlayAI product was wound down: its API went offline in late July 2025 and the platform shut down permanently on December 31, 2025.
PlayHT was started by Mahmoud Felfel and Hammad Syed, two engineers who had previously worked together at the Dubai-based online classifieds company Dubizzle (part of the OLX group). Felfel, an Egyptian engineer who earned a bachelor's degree from Mansoura University, had also worked as a software engineer at Facebook and WhatsApp before the company. The product traces back to 2016, when the founders built a Chrome extension that read Medium articles aloud, an idea that grew out of frustration with regional access to written content. The company was later incorporated in California, with most sources dating the formal founding to 2020. PlayAI was backed by the startup accelerator Y Combinator.
Over time the company shifted from a consumer read-aloud tool toward selling its underlying speech technology to developers and businesses. According to investor Tony Wang of 500 Global, Felfel and Syed were among the first founders he met, in 2022, who were using transformer models to produce human-quality conversational voice that captured the natural tones and inflections of human speech.
The company operated for years under the PlayHT name (stylized "Play.ht") before rebranding to PlayAI around the time of its 2024 funding announcement, positioning itself as "the voice interface of AI." It used the play.ai domain alongside the original play.ht address. Coverage from late 2024 onward generally refers to the company as PlayAI, while the older PlayHT branding persisted on parts of the product and documentation.
PlayAI announced a $21 million seed round on November 25, 2024. The round was led by Kindred Ventures and 500 Global, with participation from Race Capital, Y Combinator, Soma Capital, Pioneer Fund, and others. Steve Jang of Kindred Ventures joined as a board observer. The company said it would use the capital to invest in its generative voice models and voice-agent platform and to shorten the time businesses need to build human-quality speech experiences. At the time of the raise, PlayAI said it had served nearly 40,000 customers.
| Round | Date | Amount | Lead investors | Other participants |
|---|---|---|---|---|
| Seed | November 25, 2024 | $21 million | Kindred Ventures, 500 Global | Race Capital, Y Combinator, Soma Capital, Pioneer Fund |
In June 2025, Bloomberg reported that Meta Platforms was in talks to acquire PlayAI as part of an aggressive push to add AI talent. Meta confirmed the completed acquisition in July 2025 (reported by Bloomberg on July 11 and by TechCrunch on July 13). According to an internal memo reviewed by Bloomberg, the entire PlayAI team, about 35 people, was set to join Meta, reporting to Johan Schalkwyk, a former Google speech researcher who had recently joined Meta from the voice AI startup Sesame AI. The memo stated that PlayAI's "work in creating natural voices, along with a platform for easy voice creation, is a great match for our work and road map, across AI Characters, Meta AI, Wearables and audio content creation."
Financial terms of the deal were not disclosed. The transaction was widely described as a talent-focused acquisition, in line with Meta's broader 2025 spree of hires and deals (including its $14.3 billion investment in Scale AI and the recruitment of Scale's Alexandr Wang to lead a new superintelligence effort). The PlayAI team was folded into Meta's Superintelligence Labs organization.
After the acquisition, Meta did not continue operating PlayAI as a standalone service. The PlayAI API went offline on July 26, 2025, the platform stopped accepting new sign-ups in August 2025, and the service was permanently terminated on December 31, 2025. Competitors and partners publicly acknowledged the wind-down: Cartesia, a rival voice AI startup, noted in mid-2025 that PlayAI's API was already offline and that the full platform would sunset on December 31, and Groq, which had hosted PlayAI's models, fielded developer questions about the service ending. Migration guides published by competing text-to-speech vendors documented the closure, and some users reported losing access to saved audio and voice clones without a migration path.
PlayAI's offering combined hosted speech models, a voice library, voice cloning, a no-code voice-agent builder, and developer APIs and SDKs. The platform was aimed primarily at developers and businesses building voice features rather than at end consumers, and it competed with companies such as ElevenLabs, Cartesia, and the speech offerings of larger AI labs.
The company shipped several generations of deep learning speech models, which it characterized as built on the same family of large neural architectures used for large language models.
| Model | Role | Notable characteristics |
|---|---|---|
| Play 2.0 | Earlier flagship TTS model | Reported support for 900+ voices across 142 languages, with voice cloning, multi-speaker dialog, and real-time streaming |
| Play 3.0 mini | Lightweight, low-latency multilingual model | 30+ languages; tuned for real-time conversational use; reported time-to-first-byte latency around 143 ms and roughly 28% faster inference than Play 2.0 |
| PlayDialog (Dialog 1.0) | Most ambitious end-to-end speech model | Multi-turn model trained on hundreds of millions of conversations, said to be about ten times larger than Play 3.0 mini, using conversational context to control prosody, intonation, emotion, and pacing |
Play 3.0 mini was promoted as one of the fastest models of its kind, supporting text-in streaming (for example from an LLM) and audio-out streaming over an HTTP REST API, a WebSockets API, and language SDKs. PlayDialog extended the lineup with a multi-turn model designed for fluid, emotive conversation; the company said it used a conversation's historical context to make speech sound more natural than single-utterance systems. PlayDialog launched in English and later added multilingual support for languages including Chinese, French, German, Hindi, Japanese, Korean, Portuguese, and Urdu.
PlayAI let users generate speech from a large library of preset voices with adjustable intonation, cadence, and tone, and it offered voice cloning so users could create a synthetic copy of a specific voice from sample audio. High-fidelity ("instant" and professional) voice clones were a paid feature. As is common across the voice cloning industry, PlayAI's cloning capability drew scrutiny over consent and potential misuse: TechCrunch reported that the tool relied largely on users self-certifying that they had the rights to a cloned voice, with limited verification. Syed said the company had ethical safeguards in place, including mechanisms intended to identify whether a given piece of audio had been synthesized with PlayAI's technology.
Beyond raw speech synthesis, PlayAI built a no-code platform and real-time APIs for creating conversational voice agents, automated voice systems that can answer questions, handle transactions, and schedule appointments for use cases such as customer support, appointment scheduling, and sales across industries like travel, hospitality, healthcare, and retail. Reporting around the Meta deal noted that PlayAI had worked with companies including Walgreens and Salesforce on voice-agent deployments.
The company also offered PlayNote, a consumer-facing tool that turned source material such as PDFs, web pages, videos, photos, and songs into podcast-style audio, read-aloud summaries, or simulated debates. It was frequently compared to Google's NotebookLM, with PlayAI pitching the ability to use any voice, custom prompts, and API access.
PlayAI partnered with the AI inference company Groq to serve its Dialog model on GroqCloud, using Groq's hardware to deliver high-throughput, low-latency speech generation. The companies highlighted large speed gains for the Dialog model on Groq's inference stack compared with running the same model on GPUs, and the collaboration included English and Arabic endpoints. Groq's hosting of PlayAI's models became part of the post-acquisition story, as Groq users questioned the future of the Dialog and Play 3.0 mini endpoints once PlayAI began shutting down.
While the standalone service was live, PlayAI offered subscription plans for individual creators and developers along with usage-based and enterprise options. TechCrunch reported that the company's highest-fidelity voice clones were priced at $49 per month billed annually, or $99 per month billed monthly. Detailed pricing changed over the product's lifetime and ceased to be relevant after the December 31, 2025 shutdown.
PlayAI was regarded as one of the more prominent independent voice AI startups of the early 2020s, competing in a fast-growing market for synthetic speech and conversational agents. Its acquisition by Meta in mid-2025 was covered by Bloomberg, TechCrunch, Axios, and others as part of a wave of consolidation in which large technology companies absorbed specialized AI teams. The abrupt wind-down of the product, with the API taken offline ahead of schedule and the platform fully sunset by the end of 2025, left developers and creators who had built on PlayAI to migrate to alternatives, and it became a frequently cited example of platform risk in the voice AI ecosystem.