POWER
180 mAh / 12h
Estimated over 12 hours
Estimated incremental battery use over 12 hours, based on 15 mAh per hour.
HOME / PRODUCTS / VOXSENSE
VOXSENSE / ALWAYS-ON INTELLIGENCE
From sound
to intelligence.
Turn everyday conversations into useful context. Low-power audio capture and on-device transcription provide the foundation for memory and timely assistance.
01 / EXPERIENCE
Context starts with what was said.
A useful assistant needs context. VoxSense connects a continuous audio pipeline to the transcripts your product can use for memory, understanding and assistance.
01
Sense
Capture audio through a low-power path designed for extended operation.
02
Transcribe
Convert captured speech into text on the device.
03
Memory
Use transcripts to build context and personal memory in your application.
04
Assist
Connect relevant context to reminders and other timely services.
Low-power capture and local transcription are ready for product integration. Memory, intent understanding and proactive services have been demonstrated in a first demo on CAMON 40; final application behavior depends on integration.
02 / MEASURED PERFORMANCE
Always-on, with a measured power budget.
FOOTPRINT
~80 MiB
A device-sized package
Reference package including a language model. Configuration affects footprint.
12-hour estimate: 15 mAh per hour × 12 = 180 mAh. Reference setup: MTK9600, continuous speech playback, transcription once per minute. Battery impact varies by device and usage.
03 / DEVICE INTEGRATION
A local pipeline. A product you define.
Built on VoxEdge, running on the application processor. If the device has a low-power processor (e.g. SCP), it can run voice detection around the clock. Start with your chipset, audio path and transcript requirements, then connect the output to your product’s context layer.
Developer documentation ↗- Reference processing architecture
- AP, with optional low-power processor
- Low-power processor memory (optional)
- 200 KiB
- AP runtime memory
- 296 MiB
- Integration model
- SDK + licensing
Figures describe the reference test configuration.
04 / QUESTIONS
Before you integrate.
Does audio transcription run locally?
The reference audio capture and ASR pipeline runs on the device. Your application determines how transcripts are stored, used and connected to other services.
Is the full assistant experience already a product?
The capture-to-transcription solution is ready for integration. A first demo has connected audio capture, memory, intent understanding and proactive services; production behavior must be agreed for your device and application.
What should we bring to an evaluation?
Your target device and chipset, languages, microphone path, required transcript timing, and battery and memory budgets.