Cochl.Sense Nano

Cochl.Sense Nano runs sound recognition on microcontroller-class hardware—MCUs and low-power NPUs, including battery-powered devices that listen continuously. Each build is prepared for a specific chip.

1. What is Cochl.Sense Nano?

Cochl’s acoustic models are grouped by size and capability. The largest runs in the cloud. The rest run on the device itself, and Nano is the smallest of them, built for MCU-class parts that have no full operating system and a few hundred kilobytes of RAM to work with.

The set of sounds Nano recognizes depends on the chip and the build prepared for it. Tell us your target hardware and we will confirm what fits.

(1) Who is Nano for?

• Chipset vendors—you ship Nano alongside your silicon, or offer it to your own customers as part of a higher-value software bundle.
• Hardware developers—you are building a device on MCU-class parts and need sound recognition to run on it.

2. What does Nano run on?

3. What does Nano return?

Nano returns the same JSON as the Cochl.Sense Edge SDK. Each result covers a time window and lists the tags detected in it with their probabilities.

{
    "tags": [
        {
            "name": "Alarm",
            "probability": "0.757154763"
        },
        {
            "name": "Appliance_alarm",
            "probability": "0.565914631"
        }
    ],
    "start_time": "16.0",
    "end_time": "18.0",
    "prediction_time": "399"
}

start_time and end_time mark the window in seconds. prediction_time is how long the inference took, in milliseconds. As on the Edge SDK, every value is a JSON string—cast the numeric ones before comparing them.

4. How do I get Nano?

Nano is not a self-serve download. Each build is prepared for a specific chip, so the first step is a conversation about your hardware.

Email us with:

• the chip or module you are targeting
• the sounds you need to detect
• your RAM and flash budget

5. See also

• Cochl.Sense Edge SDK
• Cochl.Sense Cloud API