Hongqing Liu

Hongqing Liu

Audio Signal Processing

Speech enhancement

  • Sound quality restoration includes various aspects such as noise reduction, Howling suppression, and speech super-resolution, aiming to enhance overall audio quality.

Noise reduction

– Based on neural network solutions, non-stationary noise can be effectively removed, achieving superior results compared to traditional algorithms. The model is lightweight, making it suitable for real-time implementation.

– Before
SE before ​

– After
SE after ​

– Noise reduction DEMOPAGE.

Howling suppression

– In sound amplification scenarios, the sound emitted by the speaker is picked up again by the microphone, forming a closed loop and causing a piercing feedback noise. This not only easily damages equipment but also affects the auditory experience.

– Before
Howling before ​

– After
Howling after ​

Speech super-resolution

– The main purpose of SSR it to convert low-sample-rate speech into high-sample-rate speech to enhance the listening experience.

– Before
super before ​

– After
super after ​

– SSR DEMOPAGE.

Speech quality enhancemnet

– Interspeech2025 URGENT Challenge (Universal, Robust, and Generalizable speech EnhancemeNT systems) aims to bring more attention to constructing Universal, Robust and Generalizable speech EnhancemeNT models. To improve the universality of SE systems, the following distortions are considered: additive noise, reverberation, clipping, bandwidth extension, codec artifacts, packet loss, and wind noise.
goal ​

– 4th place globally！
results ​