Applied Brain Research Launches On-Device Speech AI SDK for Alif Semiconductor’s Ensemble Processors
The Zephyr RTOS-compatible SDK combines real-time speech recognition and text-to-speech without cloud connectivity. Running on Alif’s Ensemble E8, the Niagara ASR engine delivers 133 ms average first-token latency while consuming just 15 mW of additional active power, supporting low-power voice applications.
Applied Brain Research (ABR) announced the general availability of its speech AI software development kit (SDK) on Zephyr RTOS for the Alif Semiconductor Ensemble® family of edge AI processors. The SDK brings the most capable streaming speech recognition available, together with natural-sounding text-to-speech, running entirely on-device with no cloud connection required.
The SDK combines:
- Niagara ASR: full-vocabulary streaming speech recognition with low latency, continuously transcribing speech as it is spoken.
- Nith TTS: natural-sounding, low-latency streaming speech synthesis.
- Native Zephyr RTOS support for battery-powered and air-gapped products.
- Support for ongoing model updates and expansion.
Voice interfaces on small devices have typically depended on a cloud connection or a larger application processor. ABR's state space model technology brings capable speech AI within the power and memory budgets of microcontroller-class silicon. On the Ensemble E8, Niagara ASR achieves 133 ms for first-audio to first-token average latency while consuming only 15 mW of active power above baseline, enabling always-available voice on devices where battery life is critical.
"The Alif Ensemble family pairs the low power that edge devices demand with the AI performance to run speech workloads in real time," said Kevin Conley, CEO of Applied Brain Research. "Using these capabilities, we achieved stunning real-time performance at incredibly low power consumption. With our SDK now available on Zephyr, this is a notable milestone for what developers can do on a microcontroller-class platform."
"Zephyr is increasingly the platform of choice for our customers, and voice is one of the most requested capabilities for battery-powered devices," said Mark Rootz, Vice President of Marketing at Alif Semiconductor. "ABR's SDK gives Ensemble developers a production-ready path to on-device speech recognition and text-to-speech, and we are pleased to show it running in our booth at electronica."
Availability and demonstration
The ABR SDK for Zephyr RTOS on the Alif Ensemble family is available now. Developers can register for an account and obtain a free evaluation license at dev.appliedbrainresearch.com. Niagara ASR will be demonstrated running live on the Ensemble E8 in the Alif Semiconductor booth, C5.170, at electronica 2026, November 10–13, Messe München.
