cekal
Cekal is an individual developer and AI/ML creator, not a company or organization. His Hlas project is an expressive text-to-speech model designed to turn text into natural speech with zero-shot voice cloning and inline control of emotion, delivery, pauses, vocalizations, accents, and character voices.
At a Glance
- AI and speech researchers
- Developers building local or self-hosted text-to-speech applications
- Creators and studios prototyping expressive narration and character voices
- Interactive voice, conversational-agent, and real-time audio developers
AI Tools by cekal
(1)
Hlas
Expressive Text to Speech Model
Discussions
No discussions yet
Be the first to start a discussion about cekal
Latest News
Hlas model card and live demo link published
Hlas-1 initial release: Fortissimo fine-tune based on Breeze TTS 2
The Breeze TTS runtime and Breeze TTS 2 open-source inference stack were released, providing the documented runtime for Hlas
Products & Services
A 3.5B-parameter expressive text-to-speech model, fine-tuned on the Breeze TTS 2 architecture. It supports zero-shot voice cloning from a 3–15 second reference clip and transcript, inline performance tags and free-form directions, 24 kHz mono output, streaming inference, English and Mandarin first with a broad multilingual roadmap, and operation on a single 12 GB GPU in eager mode.
Market Position
Hlas positions itself as a compact open-weight expressive TTS and voice-cloning model: its stated differentiators are inline performance control, free-form directions, voice cloning, and a footprint suitable for a single 12 GB GPU. It is built on the Breeze TTS 2 architecture and therefore sits in the open-weight/self-hosted TTS category alongside Breeze TTS 2, while serving use cases commonly associated with hosted systems such as ElevenLabs.
Leadership
Founders
Vojtech Cekal
Independent AI/ML developer and creator. LinkedIn identifies him as CEO/co-founder of Fous and says he previously worked in security research at Google, SpaceX, and TikTok; his official site identifies him as the creator of Hlas and as building Fous.
Business Model
Revenue Model
Hlas is distributed as a downloadable Hugging Face checkpoint rather than a publicly documented paid SaaS or API product. The model weights are under the BreezeBlue Research and Non-Commercial license; the inference runtime is Apache-2.0.
Target Markets
- AI and speech researchers
- Developers building local or self-hosted text-to-speech applications
- Creators and studios prototyping expressive narration and character voices
- Interactive voice, conversational-agent, and real-time audio developers
- Expressive speech synthesis for interactive applications
- Zero-shot voice cloning from short reference recordings
- Audiobooks, narration, and character voices
- Dialogue and game or animation voice prototyping
- Voice assistants and conversational agents requiring emotion, pacing, or vocal events
- Local and self-hosted research experimentation with open-weight TTS