- Semiconductor BusinessHOME
- Products and Services of Macnica,Inc.
-
technical information
-
Events and Seminars
- Handling Manufacturer
- Support
- Inquiry
- Click here to purchase products
- Semiconductor business e-mail magazine registration
Narrow down by specifying conditions
現在2209件がヒットしています。check
This article presents a voice recognition AI (Voice AI) demo utilizing the Astra ™ SL2610 AI processor, which was showcased at Embedded World 2026.
In recent years, while the use of generative AI has advanced rapidly, there has also been an increase in cases where challenges such as constant connectivity to the cloud, response delays, and data handling become apparent.
The Astra SL2610 introduced here is an edge AI processor developed to address such challenges.
A key feature is its Coral NPU, developed in collaboration with Google Research, which enables the execution of Transformer-based AI models with low power consumption.
Voice AI at the Edge with Synaptics Coral Dev Board and Astra SL2610
Demo of running voice AI at the edge
The video demonstration showed a process where the user's spoken voice was input via microphone, the audio segment was extracted using VAD, Moonshine converted it to text, and Gemma 3 understood the content and generated a response.
The Moonshine speech recognition model used is a Transformer-based speech-to-text model. It transcribes user speech into text in real time and passes it on to the next AI processing stage. The generated text is input into Gemma 3, Google's open-source LLM, which generates answers tailored to the question.
The demo showed the device instantly responding to the question, "Tell me about Synaptics Astra."
By performing these processes on edge devices rather than in the cloud, we are able to reduce response delays and ensure privacy protection.
Edge AI development environment provided by the Astra platform
This demo not only showcases the performance of the Astra SL2610, but also demonstrates the potential of the Astra platform developed by Synaptics.
Astra is an integrated hardware and software development platform for edge AI. It provides a comprehensive environment that includes Astra series processors, development tools, AI frameworks, and a partner ecosystem.
Another key feature is its support for open-source AI. Developers can leverage existing open-source models to develop AI solutions tailored to their specific applications. It is positioned as a platform that encompasses a wide range of application development, including not only voice AI, but also image recognition, video analysis, and multimodal AI.
Expected applications and future possibilities
Edge Voice AI is a technology that is very well suited to environments where constant connectivity to the cloud is difficult, and to devices that want to process data locally. From this demo, we can expect it to be used in the following applications:
・AI home appliances
Smart Home Hub
Smart City Terminals
Furthermore, Synaptics offers the Astra series for a wide range of IoT applications, including consumer electronics, industrial equipment, and enterprise devices.
As the use of generative AI expands, Voice AI that can be completed entirely at the edge is expected to become increasingly important. The Astra SL2610 is attracting attention as a new generation edge AI processor that will support the realization of this.
Summary
The Astra SL2610 combines speech recognition powered by Moonshine with AI processing utilizing Gemma 3 to create a cloud-independent edge AI environment.
Because processing can be completed on the edge side, it provides a comfortable user experience with low latency, and since audio data can be processed without being sent externally, it also offers significant advantages in terms of privacy protection and security.
This demo, unveiled at Embedded World 2026, demonstrates the potential of edge AI applications, including speech processing, and is expected to be deployed in a wide range of applications.
The Astra platform enables development using open-source AI models.
For example, communities such as GitHub have published samples that use small language models like Gemma 3 270M to translate speech recognition results into multiple languages, and applications to speech translation and multilingual interfaces are expected.
Link: https://github.com/synaptics-astra-demos/sl2610-examples
We should also keep an eye on the future evolution of the edge AI platform being promoted by Synaptics.
