How an AI note taker device works: recording, Bluetooth transfer and transcription
An AI note taker device works in 3 stages. The device records and compresses audio locally on the hardware. It then transfers the audio file to a smartphone or computer via Bluetooth Low Energy. Finally, a companion app or cloud service transcribes the audio into text. The device itself only handles recording and storage.
What is an AI note taker device
An AI note taker device is a dedicated hardware recorder that captures audio, stores it locally, and relies on a connected smartphone or computer to handle transcription and cloud synchronization via a documented wireless protocol.
Unlike a traditional voice recorder that simply saves a file, this device is designed to be the initial step in a larger software pipeline. It captures the raw acoustic data efficiently and hands it off to a more powerful processor, like a smartphone, to do the heavy computational work of speech recognition. You can see a practical implementation of this hardware in our AI recorder product line.
Stage 1: Recording and audio compression
The device uses microphones to capture sound and immediately compresses it using the Opus codec. Opus is highly efficient for speech, allowing the device to store long recordings without filling up local flash memory.
On our test hardware, the device records in mono at approximately 16 kbps. The audio is broken down into fixed 40-byte packets, with each packet representing 20 milliseconds of audio. This structure is defined in the Opus RFC specification for low-delay communication.
Because of this compression rate, a continuous 1-hour recording results in a file size of about 7.2 MB. In our laboratory tests, a file with a duration of 60:01 measured exactly 7202040 bytes. This small footprint is critical because the device must transfer this data over a low-bandwidth wireless connection later. For a deeper technical breakdown of file sizes, see our article on AI voice recorder data size and transfer time.
Stage 2: Bluetooth Low Energy transfer
Once the recording is saved, the device transfers the file to a host device using Bluetooth Low Energy (BLE). BLE is chosen over Wi-Fi because it consumes significantly less power, allowing the device to maintain a standby time of several weeks on a small battery.
Transfer speeds vary heavily depending on the host operating system and its BLE stack implementation. The device sends data in 160-byte notification packets. The host must acknowledge these packets, and the speed is bottlenecked by the connection interval and the Maximum Transmission Unit (MTU).
| Host Device | Test File | Transfer Time | Average Speed |
|---|---|---|---|
| Android (Xiaomi Mi 6, native app) | 1-hour recording | 3 min 49 s | 30.7 KB/s |
| iPhone 16 (third-party app) | 1-hour recording | About 4 min | 28 to 33 KB/s |
| Windows 10 (Python native BLE) | 1-hour recording | About 22 min | 2 to 18 KB/s (fluctuating) |
On mobile platforms, the BLE stack can negotiate a fast connection interval of 15 ms and an MTU of 247 bytes, allowing the device to push data at roughly 30 KB/s. On Windows 10, the operating system restricts applications from modifying the connection interval, leading to slower and highly fluctuating transfer speeds.
Stage 3: Transcription and device limitations
The device itself does not run speech-to-text algorithms. It has no operating system capable of hosting large neural networks, nor does it have a direct internet connection to call cloud APIs. Its only job is to record and store.
Transcription happens entirely on the companion app or a connected computer. The app reads the raw Opus file from the device, wraps it in an Ogg container so standard media players can decode it, and then sends the audio data to any OpenAI-compatible speech-to-text service with your own key, or a local model you host.
Live audio streaming over BLE (pushing 160-byte packets while recording) is available in the order edition of the reference software, typically from 1,000 units. The free and sample editions download finished files after recording, then transcribe them in the app. The app buffers these packets and sends them to the transcription API in chunks, allowing the user to see text appear on their phone screen while the meeting is still happening. The device continues to save the complete, high-quality file locally for a more accurate batch transcription later.
What the open Bluetooth protocol allows
For brands integrating this hardware, the device exposes a documented BLE protocol. This allows a custom app to control the hardware without needing proprietary SDKs. The protocol supports the following commands:
- List all recorded files with their duration and storage size.
- Download files entirely or in segmented chunks for resumable transfers.
- Delete specific files or clear the entire storage.
- Remotely start, stop, and pause the recording process.
- Read device status, including battery level, remaining storage capacity, and firmware version.
- Adjust microphone gain settings.
- Order-edition builds can receive a live stream of audio data for real-time processing.
This open architecture means the hardware can be white-labeled and integrated into any existing software ecosystem.
Design trade-offs for brands building a device
When brands decide to build or source an AI note taker, they must balance several engineering constraints. The choice of BLE over Wi-Fi is a deliberate trade-off. Wi-Fi would offer much faster transfer speeds, but it would drain the battery in hours rather than weeks. For a wearable or pocket-sized device, battery life is the primary user requirement.
Another trade-off is the audio format. Storing raw WAV files would require 16 times more storage space, filling up a 32 GB card in just a few hours. Using Opus at 16 kbps solves the storage problem but requires the companion app to handle the decoding and container wrapping.
Brands also need to consider the software development effort. The hardware is only half the product. The real value is in the companion app and the transcription pipeline. If you are evaluating manufacturing partnerships, understanding the difference between a full custom build and leveraging existing reference designs is critical. You can read more about these manufacturing choices in our guide on ODM vs OEM AI hardware.
FAQ
Does the device transcribe audio on its own?
No. The device only records and stores audio. Transcription is handled by the companion app on a smartphone or computer, which sends the audio to a speech recognition service.
How long does it take to transfer a 1-hour recording?
On a smartphone, it takes about 4 minutes. On a Windows 10 computer, it can take up to 22 minutes due to operating system limitations on Bluetooth connection parameters.
What audio format does the device store?
The device stores raw Opus audio packets at 16 kbps. A 1-hour recording is about 7.2 MB. The companion app must wrap these packets in an Ogg container for playback.
Can the app control the recording remotely?
Yes. The Bluetooth protocol allows the companion app to remotely start, stop, and pause the recording, as well as adjust microphone gain and read the battery level.
Is Wi-Fi required for the device to work?
No. The device uses Bluetooth Low Energy to transfer files to a smartphone. The smartphone then uses its own internet connection to send the audio to a cloud transcription service.
About this page
This page is published by PINGOOD TECHNOLOGY CO. LIMITED, The content applies to our current generation of Bluetooth Low Energy AI note taker devices. The transfer speeds and file sizes mentioned in this article are measured data from our laboratory tests on firmware version 1.0.0. The calculation of the 7.2 MB file size is derived from the measured 16 kbps bitrate over a 1-hour duration. No theoretical maximums or unverified claims are included in this text.
Related
Want to try the hardware behind these notes?
Six AI hardware platforms are in mass production. Tell us your market and channel and we send an evaluation sample.
Request a sample