by Sree Durbha, Senior Director of Product Marketing at Alif Semiconductor
The next frontier for wearable audio is to incorporate artificial intelligence.
Compact wearable audio devices such as earbuds and hearing aids are powered by microcontroller-based systems-on-chip (SoCs), which are marvels of engineering with their high integration. These SoCs integrate general purpose computing and high- performance digital signal processing (DSP) capabilities, which handle crucial audio functions—such as codec processing, noise reduction, FFT analysis, and compression—alongside Bluetooth network functionality, battery management, and system control. This level of integration not only reduces the bill of materials but also enables manufacturers to deliver superior audio quality in tiny, wearable formats.
However, the game is about to change. The next frontier for wearable audio is artificial intelligence (AI). By embedding AI directly into the SoC, manufacturers can dramatically enhance audio functions such as noise cancellation and echo cancellation, while also introducing advanced features such as speech recognition and natural language processing (NLP) for key word spotting (KWS).
The challenge lies in implementing AI without compromising the compactness, power efficiency, or cost of these devices.
The power of AI in wearable audio devices
AI has the potential to revolutionize the wearable audio market. Traditional noise cancellation, called active noise cancellation (ANC), relies on fixed algorithms that continuously process ambient noise and emit an opposite-phase signal to cancel the ambient noise signal: this method consumes substantial amounts of power while struggling to adapt to varying environments.
AI, however, goes about noise cancellation in a different and less power-hungry way. By analyzing ambient noise and identifying its unique signature, AI can dynamically select the most effective noise cancellation algorithm from a library of pre-trained models, adapting in real time to different surroundings. This AI-based technique does not require continuous active listening to the environment for ambient noise through multiple microphones, thereby saving a lot of power.

Several companies are already leading the charge with off-the-shelf AI software for noise cancellation, providing offerings called enhanced noise cancellation or AI noise cancellation.
The advantages are clear: AI-enhanced noise cancellation not only adapts more effectively to changing aural conditions, but also significantly reduces power consumption by enabling periodic noise sampling. This innovation extends the battery life of wearable devices, providing for longer use without sacrificing performance.
The integration of AI into other DSP functions—such as speech processing, KWS, and echo cancellation—offers similarly transformative benefits.
Overcoming the limitations of traditional SoCs
Attempting to run this type of audio AI model on the traditional DSP-rich SoCs used in wearable devices, however, often results in subpar performance. DSPs and CPUs such as the Arm Cortex-M7 core are optimized for sequential operations, making them ill-suited for the highly parallelized computing required by AI’s neural networks. This mismatch leads to high power consumption and slow response times, particularly in time- sensitive functions such as KWS.
Figure 1 illustrates the stark difference in performance between a standard CPU and a combination of a CPU and neural processing unit (NPU) when executing common AI tasks. The comparison underscores the need for a dedicated NPU optimized for AI tasks, such as the Arm Ethos NPU core, which excels in edge AI devices due to its ultra-low power consumption and seamless integration with the Cortex-M CPU family (M55 and beyond) .
Architecting SoCs for AI excellence
Enhancing a wireless audio SoC with AI capabilities involves more than just adding an NPU. To fully harness the potential of AI while maintaining power efficiency, the SoC must feature ample tightly coupled memory (TCM) and advanced power management that can selectively power down portions of the MCU when not in use. The choice of CPU core is also crucial. For instance, the Cortex-M55 core, with its Arm Helium M-profile vector extension (MVE), offers substantial improvements in machine learning
(ML) and DSP functions, outperforming even the high-end Cortex-M7 core and offering up to 4x better ML performance and 3x better DSP performance.

AI performance sees its most significant boost when an NPU is added. Figure 2 showcases the performance benefits of an Alif Ensemble MCU that combines dual Cortex-M55 cores with dual Ethos-U55 NPUs, delivering outstanding results in running inferencing on ML models, including one for KWS on-device in the endpoint.
Specifically for wireless audio device manufacturers, the benefit of choosing a Cortex-M55-based SoC with Ethos NPU is revealed in test results for the crucial AudioMark benchmark specified by EEMBC. According to EEMBC, “The AudioMark Benchmark is the first-of-its-kind audio benchmark that incorporates advanced signal processing, multiple data types, and a convolutional neural net in a single benchmark with a realistic code footprint.”
This benchmark is intended to measure performance in AI functions such as KWS as well as traditional audio processing functions such as beam forming along with echo and noise cancellation. As Figure 2 shows, the combination of the Ethos-U55 NPU and Helium MVE working alongside the Cortex-M55 core provides a substantial uplift in benchmark performance compared to the Cortex-M55 executing the inference by itself.
Pioneering AI-optimized SoCs
A prime example of this AI-optimized architecture is Alif Semiconductor’s Balletto B1 MCU. Housed in a compact package, the Balletto B1 integrates a Cortex-M55 CPU, an Ethos-U55 NPU, and up to 2MB of tightly coupled SRAM. Its 2.4GHz radio subsystem supports Bluetooth Low Energy v5.3 and 802.15.4 networking, making it a powerful yet efficient solution for AI- driven wearable audio devices. Balletto B1 outperforms existing DSP cores in both AI tasks and Bluetooth operations, setting a new standard for the industry.
The future of wearable audio
The introduction of AI-optimized wireless MCUs such as the Balletto B1 signals a new era in wearable audio. As demand for AI capabilities in small, power-efficient devices grows, the market is responding with innovative solutions which enable manufacturers to develop compact, high-performance products with extended battery life.
The integration of AI into wearable audio SoCs is not just an upgrade, it’s a revolution that promises to redefine the listening experience.