Released AudioDecisionModel, a Jev-style Audio Decision AI

mocomoco Inc. has released AudioDecisionModel, a Jev-style audio decision AI, with open weights.

It answers questions about short audio clips with Yes/No, a choice, or a number. You can ask about speech, as well as music, animal sounds, and ambient noise. An input and output example looks like this.

Example Questions about audio, and the answers

Input

Audio: A short clip that includes a dog barking Sample audio. Amada44 / CC BY-SA 3.0
Context: Check whether a dog’s voice is in the clip

Questions JSON

{
  "Is there a dog in the audio?": {
    "type": "noul",
    "instructions": "Check whether the audio includes a dog barking."
  }
}

Output

Answers JSON

{
  "Is there a dog in the audio?": "Yes"
}

mocomoco inc. holds the mission “Gathering human voices to circulate as true value.” Those voices have to reach the decisions and actions that need them, so the model also has to run in the field and on the device in hand. With that in mind, we released this model as a contribution to in-browser edge execution and to more accessible open-weight models.

The model and demo are on Hugging Face.

This model draws on Jev from TypeSafe AI. The base models it uses are:

For inquiries about this announcement, please use our contact form.