Pioneering Edge AI for Unparalleled Audio Experience
Are you seeking to elevate your audio products with pioneering AI algorithms? Do you wish to provide your customers with crystal-clear sound quality and immersive audio experiences? Discover the potential with CrystalSound.AI, the leading provider of real-time speech enhancement solutions designed specifically for edge devices.
Where the innovation meets quality
With CrystalSound.AI, benefit from our wide range of state-of-the-art AI algorithms that enhance the performance and functionality of your audio devices. Our solutions are designed to deliver low latency, minimal speech distortion, and real-time speech enhancements.
Wide Range of Platform Support
Embracing Diverse Platforms with CrystalSound.AI
Our cutting-edge algorithms are designed for flexibility and integration. They can operate seamlessly across a broad spectrum of platforms, from bare metal to desktop OS, and can be easily incorporated into your existing hardware and software infrastructure.
Bare Metal Support
Whether running on DSP or CPU, our algorithms have been tested and proven to perform optimally on a variety of platforms, including:
- DSP: Specifically designed for Cadence HiFi DSP 4 and 5 (tested on Airoha 1585 and NXP i.MX RT600).
- CPU: Compatible with ARM Cortex A53 and up, catering to most ARM SOCs from MediaTek, Qualcomm, and NXP. Our algorithms also run smoothly on Intel and AMD CPUs.
Embedded OS Support
CrystalSound.AI algorithms can be seamlessly integrated with various embedded operating systems, including:
- Yocto Linux
- Embedded Android
- Windows for IoT
Desktop OS Support
Our technology is compatible across major desktop operating systems including Windows, MacOS, and Linux, with dedicated support for Windows APO.
We offer full support for WebAssembly, enabling more effective and secure execution on users' web browsers. This means, you can use CrystalSound.AI to improve speech quality on your web front-end, without any server cost or hassle.
CrystalSound on Cadence's HiFi DSP 4 and 5
Superior Performance and Efficiency with CrystalSound
Our solutions are designed for efficiency and superior performance. Our real-time noise reduction outperforms competitors, with significantly lower speech distortion, using only a fraction of the computational resources.
CrystalSound Performance on Edge
Winning the Edge Performance Race
Our 3rd generation model showcased stellar performance, outperforming traditional WebRTC noise reduction and RNNoise. It also demonstrated significantly less speech degradation than other AI-based speech enhancement providers.
P835 DNSMOS performance on blind test set of Microsoft' deep denoise challenge 4 (2022)
CrystalSound Performance on Edge vs Desktop
Bridging the Gap: Edge vs Desktop
On Airoha devices, our cutting-edge model was able to achieve 75% of the performance of the Desktop model, with a 25x reduction in size.
Driven by Feedback and Innovation
Our success lies in the continuous refinement of our algorithms, fueled by the feedback from our growing user base of PC app users. Our proprietary speech dataset features more than 100,000 unique speakers across 99 languages, which helps us refine and improve our algorithms. We also value our clients’ privacy and do not use their data for training. Instead, we rely on public datasets with permissive licenses and conduct our own data collection with consent.