cbbh/hviske-v3-coreml

Hviske v3 conversation (WhisperKit Core ML)

This repository contains Core ML artifacts compatible with WhisperKit for the Danish ASR model:

• Source model: syvai/hviske-v3-conversation
• Runtime: WhisperKit (Core ML)
• Platform: macOS (Apple Silicon)

Converted artifacts notice: This repository contains Core ML (.mlmodelc) artifacts converted from the Hugging Face model syvai/hviske-v3-conversation (Licensed Material). These files are Adapted Material under the license. Conversion performed by Christian (cbbh) on 2026-02-09 using WhisperKit/whisperkittools. All use and redistribution must comply with the full terms in LICENSE, including the use-based restrictions, and the non-commercial/commercial terms stated by syv.ai.

What’s inside

The Core ML models are stored under:

• syvai_hviske-v3-conversation_large-v3/

and include:

• MelSpectrogram.mlmodelc
• AudioEncoder.mlmodelc
• TextDecoder.mlmodelc

These are intended to be downloaded by WhisperKit using a modelVariant of large-v3 (WhisperKit searches for *large-v3/* in the repo).

License

The original model syvai/hviske-v3-conversation is licensed by syv.ai.

• Non-commercial use: OpenRAIL-style license (see LICENSE)
• Commercial use: contact mads@syv.ai

This repo redistributes converted artifacts derived from the Licensed Material, and inherits the same use restrictions.

Use restrictions (high level)

You must comply with the restrictions in LICENSE, including (non-exhaustive):

• Do not impersonate a person or create synthetic speech emulating a specific natural person.
• Do not use the model to detect or infer identity attributes of natural persons.

WhisperKit usage (Swift)

Typical initialization (download on first run):

import WhisperKit

let config = WhisperKitConfig(
  model: "large-v3",
  modelRepo: "cbbh/hviske-v3-coreml",
  modelToken: nil,
  download: true,
  load: true,
  prewarm: true
)

let whisperKit = try await WhisperKit(config)

Attribution

• syv.ai: model fine-tuning and licensing
• WhisperKit: Core ML runtime