In May 2024, Microsoft defined the hardware and software specifications for AI PCs. Among the various technological applications that followed, the “AI noise reduction technology” has captured the public’s imagination. Unlike traditional noise reduction techniques, AI noise reduction is theoretically capable of more accurately distinguishing between speech and background noise, even identifying critical alarms from general noise. However, as this technology advances, it also faces new challenges in functional and performance verification.
Challenges Faced by the Client
The new AI noise reduction technology can be applied not only to microphone noise reduction but also to filtering out background noise from the other party during video conferences. This allows the local speaker to deliver a clearer voice from the other side. As this technology progresses, manufacturers must evaluate noise reduction performance considering various noise sources and application scenarios.
In traditional noise reduction performance evaluations, pink noise or white noise is typically used as the test noise source. However, real-world environments are not composed of such pure noise; instead, they consist of a variety of environmental sounds such as surrounding voices, music, vehicles, fans, and random noises generated by the operation of devices. A well-known domestic laptop brand sought to understand the performance of AI noise reduction in its AI PC laptops in real-world environments. Since the brand did not have a real-world noise sound field setup, they sought assistance from Allion Labs’ consulting team to determine if there was a corresponding solution to measure the AI noise reduction performance.
Allion Labs’ Solution
With the AI noise reduction technology able to recognize different types of noise, traditional evaluation methods are no longer sufficient. To address this, Allion Labs invested heavily in 2023 to build a sound field reproduction technology that can realistically recreate various environmental noises, thus providing a more accurate testing environment
Reference：Ambient Noise Sound Field Restoration Test Lab | Allion Labs
After understanding the use cases for this AI PC, the Allion Labs consulting team developed a validation plan to simulate real consumer scenarios. The key points include:
- Reconstructing the noise environment of a café.
- Using AVTP (Allion Voice Test Platform) for intelligent, automated voice measurement.
- Performing multiple fully automated repeated tests to obtain average data.
The product’s performance for features like “microphone AI noise reduction” and “speaker AI noise reduction” was measured, comparing success rates of voice recognition in both enabled and disabled states over multiple tests
The measurement results indicate that the AI noise reduction technology in this AI PC significantly improves speech recognition accuracy:
- Enabling microphone AI noise reduction improves accuracy by 5.4%.
- Enabling speaker AI noise reduction increases accuracy by an average of 16.8%.
Time to Market with Quality! High-Quality Service to Create Value for You
Faster, Easier, Better! Allion Labs Team
Allion Labs delivers Faster, Easier, and Better! From the initial client inquiry to planning the validation process, executing the tests, and delivering the report, the Allion Labs consulting team completed the entire project in just one week. We provided clients with diverse value through the Faster, Easier, Better approach:
- Saving time and costs on setting up application environments.
- Simulating and reconstructing real-world environmental noise.
- Offering precise data through automated measurements.
- Providing long-term and repeated measurement data.
- Reducing manpower requirements
Allion Labs’ acoustics consulting team has established advanced professional equipment to offer expert audio and electroacoustic measurements, as well as development and design consulting services. The above measurement solutions and experimental result analysis allow for a concrete evaluation of the AI noise reduction performance of microphones and speakers in AI PCs under specific real-world noise environments. Additionally, the detailed technical analysis report helps shorten product development timelines, achieving the goal of Time to Market with Quality!
If you have any questions or requirements, please contact us.