You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this model content.

Kiri-S2S β€” Real-Time Khmer Speech-to-Speech AI

Kiri (αž‚αžΈαžšαžΈ) is an end-to-end, real-time speech-to-speech conversational AI model built Khmer-first by Blizzer.

Speak to it in Khmer and it talks back in a natural voice in under a second β€” streaming audio while the rest of the reply is still being generated. It also converses in English and supports live translation.

The name "Kiri" comes from the Khmer word for "mountain" (αž‚αžΈαžšαžΈ).

πŸ‘‰ Try the Live Demo: s2s.blizzer.tech
πŸ“– Read the Launch Post: English Blog | αž’αžαŸ’αžαž”αž‘αž‡αžΆαž—αžΆαžŸαžΆαžαŸ’αž˜αŸ‚αžš (Khmer)


Key Capabilities

  • Speech-to-Speech (S2S): Understands raw audio input directly and generates spoken responses natively without separate intermediate conversion steps.
  • Real-Time Streaming: Responds in <1 second regardless of reply length; starts speaking while processing continues.
  • Khmer-First & Bilingual: Native Khmer and English support, seamlessly switching languages or translating live on request.
  • 4 Built-in Native Voices: Select between Bosba, Borey, Setha, and Maly.
  • Live-Call Ready: Designed for real-time streaming voice interactions with barge-in support (interrupt mid-sentence).

Model Overview

  • Model Size: 2.04 Billion Parameters
  • Architecture: End-to-End Native Speech-to-Speech Neural Architecture
  • Tokenizer Efficiency: Extended with 4,000 native Khmer tokens (βˆ’80% token processing cost for Khmer compared to standard multilingual models).

Current Limitations (Research Preview)

  • Research Preview: May occasionally make factual errors, misinterpret background noise, or produce unnatural phrasing.
  • Language Scope: Optimized for Khmer and English only.
  • Voice Options: Fixed to the four built-in speakers.
  • Non-Professional Advice: Not intended for medical, legal, or financial decision-making.

About Blizzer

Developed by Blizzer, a Cambodian AI startup dedicated to building voice AI technology that natively serves Khmer speakers.

Downloads last month
20
Safetensors
Model size
2B params
Tensor type
F32
Β·
BF16
Β·
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support