How Does Call Noise Reduction Work on a Single-Microphone Phone?

Single-microphone phones face unique challenges when it comes to noise reduction compared to multi-microphone devices. Since there is only one microphone to rely on, it becomes difficult to differentiate between speech and background noise based on spatial information. The only way to reduce noise on these phones is through inherent platform processing, which includes noise suppression, gain control, limiting, and echo handling.

These features are tuned separately for handheld and speakerphone modes, with the ultimate evaluation criterion being how the person on the other end of the call perceives the conversation. However, the single-microphone problem presents several challenges that can impact call quality. With only one microphone, there is no way to separate overlapping speech signals, making it difficult to distinguish between multiple speakers.

This makes it impossible to filter out background noise effectively. Additionally, aggressive noise suppression can accidentally eliminate important speech elements, leading to a distorted conversation for both parties. The article breaks down the single-microphone noise reduction process into three capability tiers. The basic tier focuses on platform tuning and tailoring the native call audio path for each microphone mode.

This results in some improvement in call quality, but there are still limitations in handling steady noise and transient sounds. The advanced tier introduces lightweight scene processing, which allows for adaptive suppression strength and weak-speech protection. This leads to better intelligibility in noisy environments, but requires additional compute resources and verified hardware capabilities.

The high tier, which involves complex-scene enhancement, is still a research direction and not a current delivery commitment. The article also highlights seven real-world scenarios that determine call quality, including steady noise, wind, transients, soft speech, speakerphone echo, and more. Each of these scenarios has specific failure modes, and the basic tier tuning focuses on addressing the most common issues.

The full validation method for the single-microphone noise reduction process is published in the PMAOS evidence repository, allowing for thorough testing and evaluation before implementation.