Models · Alebex
Models

Voice AI, built in Vancouver.

Beyond the platform, the models are callable directly over the API — the way you would call any other model provider. Speech-to-text arrives first; text-to-speech follows as coming soon.

Alebex Voice Engine
The runtime
Speech-to-text
Ships first
Text-to-speech
Coming soon
Alebex Voice Engine — the stream, in order. Illustrative
Telephony·8 kHz mulaw over media streams
Browser voice·16 kHz PCM
01
Audio in
02
Noise reduction
03
Voice activity detection
04
Streaming speech-to-text
05
Turn control
06
LLM
07
Streaming text-to-speech
08
Audio out
Per frame
Turn end + barge-in
Token stream, gated
Per sentence
Interruptible

One pipeline serves both, with zero code forks.

Why these models exist

Most systems — wrappers
around the same pipeline.

Most voice AI products are assemblies of the same pipeline: third-party speech-to-text, an LLM, third-party text-to-speech, telephony, a dashboard. Alebex builds the stack instead.

The orchestration platforms assemble third-party models; Alebex's quality mechanisms exist because it owns the layers they would have to bolt on.

The stream, stage by stage Illustrative
Step Stage Detail
01Audio in—
02Noise reductionPer frame
03Voice activity detection—
04Streaming speech-to-text—
05Turn controlTurn end + barge-in
06LLMToken stream, gated
07Streaming text-to-speechPer sentence
08Audio outInterruptible
Alebex Voice Engine
Catalogue

The engine, and the models
it is made of

One runtime underneath all of it. The engine is what the platform runs on; the models are the layers it owns, arriving on their own as they are ready.

The runtime
Alebex Voice Engine

A real-time voice engine where everything streams end to end: streaming speech-to-text, turn control, the hallucination gate, barge-in classification, multimodal intent understanding and streaming synthesis.

Stream Illustrative
01Audio in
02Noise reduction
03Voice activity detection
04Streaming speech-to-text
05Turn control
06LLM
07Streaming text-to-speech
08Audio out
Ships first
Speech-to-text

Context-aware linear streaming transcription: full-context transcription at linear cost, any turn length.

Transcribing Illustrative
turn 01
I need to change the intake date on my application.
turn 02
Full context held for the whole turn, no window resets…
turn 03
Can you confirm the new date once it is updated…
Coming soon
Text-to-speech

Streaming synthesis, per sentence, with interruptible audio out.

Coming soon
Synthesis Illustrative
Models pricing Coming soon

Model pricing is coming soon.

Model pricing is published once go-to-market pricing is established. It is listed in the navigation and marked coming soon until then.

Models pricing
The developer motion

Build with our API

Build with our API
Build with our API

Voice sessions, streaming and call control through open REST and WebSocket APIs, on a self-serve developer account.

Session request Illustrative 
POST /v1/voice/sessions
Authorization: Bearer sk_live_…

{
  "model": "alebex-voice-engine",
  "transport": "telephony",      // 8 kHz mulaw
  "audio_out": "pcm",             // 16 kHz
  "stream": true
}
Get a live demo

Ready to hear the agent call you?

Drop in your website and Alebex reads it, learns what you do, and calls you in under 30 seconds.

Call me now