KittenML
KittenML (the product brand of Stellon Labs Inc.) builds tiny, efficient speech and other AI models for edge devices. Its current focus is expressive text-to-speech and emotion-aware speech recognition that can run on CPUs, consumer GPUs, phones, wearables and embedded systems.
At a Glance
- Developers and AI application builders
- Software and hardware companies deploying local AI
- Voice-agent and contact-center teams
- Media/content and game developers
- +3 more
AI Tools by KittenML
(1)
Kitten-TTS
Open Source Text to Speech Models
Discussions
No discussions yet
Be the first to start a discussion about KittenML
Latest News
Independent Medium feature highlights KittenTTS's compact CPU speech models, voice cloning and expressive controls.
KittenTTS repository merged consumer-GPU support work.
KittenTTS prepared version 0.9.7 and refreshed GPU compatibility and vLLM documentation.
KittenTTS added an MLX backend for faster Apple-silicon inference.
Products & Services
A 1.7B-parameter ternary speech-language model for expressive text-to-speech, voice cloning from short reference audio, streaming/chunked generation, multilingual speech and emotion, vocal-event and emphasis controls. It runs on CPUs and consumer GPUs and is available for self-hosting and through the hosted API.
Lightweight open-source ONNX TTS models: mini (80M), micro (40M) and nano (15M; an int8 build is about 25 MB), designed for real-time CPU/on-device inference.
Hosted speech-to-text with emotion-aware/enriched transcripts that can preserve delivery information such as tone, stress, pauses, excitement and other markers; supports upload and realtime streaming.
A web playground and OpenAI-compatible API at api.kittenml.com/platform.kittenml.com for TTS, voice creation/cloning, transcription and realtime inference.
Market Position
KittenML positions itself as a highly efficient, low-cost and deployable alternative to cloud-first voice APIs, emphasizing CPU/edge operation, small model footprints, open/self-hosted weights and expressive controls. In lightweight/open TTS it competes with Piper, Kokoro, Bark and Coqui XTTS; in hosted voice APIs it benchmarks against ElevenLabs, MiniMax, Cartesia and Deepgram. Its differentiation is efficiency and local deployment rather than maximum raw voice realism of the largest cloud models.
Leadership
Founders
Rohan Joshi
Co-founder and CEO; former Meta Research engineer/researcher and Carnegie Mellon School of Computer Science alumnus.
Divam Gupta
Co-founder and CTO; Carnegie Mellon robotics alumnus, former Microsoft Research and Meta AI/Research engineer, and creator of the one-click ML tool Liner.ai.
Executive Team
Rohan Joshi
Co-founder and CEO
Former Meta Research engineer/researcher and Carnegie Mellon SCS alumnus; leads the company and its open-source frontier speech-model effort.
Divam Gupta
Co-founder and CTO
Former Microsoft Research and Meta AI/Research engineer with Carnegie Mellon robotics training; creator of Liner.ai and focused on compact AI models.
Founding Story
Rohan Joshi and Divam Gupta started Stellon Labs after working on AI research and codec-avatar models at Meta and Microsoft/Carnegie Mellon. They wanted to address the compute and memory demands that keep conventional foundation models off phones, wearables, robots and embedded devices, beginning with a sub-25MB open-source TTS model and expanding toward tiny frontier models for speech, language and video.
Business Model
Revenue Model
Hosted API usage, monthly subscriptions, pay-as-you-go inference, enterprise/private deployments, custom capacity, custom fine-tunes/voices and commercial support/licensing. Self-hosted Kitten-TTS 2 commercial use is free for companies under the stated $1M annual-revenue-and-total-funding threshold; larger commercial self-hosting requires a paid agreement.
Pricing Tiers
30 free minutes, then pay as you go; TTS 2 $15 per 1M characters and ASR Pro $1 per audio hour; includes voices, languages and voice cloning.
$24.99 of monthly usage included; TTS 2 $0.65 per 1M characters and ASR Pro $0.10 per audio hour.
Private deployment, volume pricing, custom usage/rates, custom API capacity, invoicing and dedicated support.
Target Markets
- Developers and AI application builders
- Software and hardware companies deploying local AI
- Voice-agent and contact-center teams
- Media/content and game developers
- Accessibility and education organizations
- Robotics, smartphone, wearable and embedded-device makers
- Voice agents
- Customer-support call automation and triage
- Content creation, narration, dubbing and audiobooks
- Accessibility and assistive applications
- Games and character voices
- Education