SEND: A Spatial Event Neural Detector for Intentional Object Motion in Immersive Music Mixing

Xu Gan,Linhao Zhao,Zhenhai Yan

Tencent Music Entertainment

We present SEND,(Spatial Event Neural Detector), a dual-stream architecture that mimics an engineer’s holistic listening by decoupling audio into a target stream (the dynamic object) and a background context stream (the static accompaniment mix).

⚠️ Please use high-quality headphones for the best spatial listening experience.

Model Architecture & Flow

Model Architecture & Flow — Dual-stream architecture: a content stream encodes spectral context, while an event stream detects intentional motion cues.

Audio Samples

Best with headphones

1

Demo1

Pop

Anchor

Original stereo mix

Reference

Human expert mix

DSP

DSP baseline result

CRNN

CRNN baseline result

SEND (Ours)

Our proposed model result

2

Demo2

Latin

Anchor

Original stereo mix

Reference

Human expert mix

DSP

DSP baseline result

CRNN

CRNN baseline result

SEND (Ours)

Our proposed model result

3

Demo3

Rock

Anchor

Original stereo mix

Reference

Human expert mix

DSP

DSP baseline result

CRNN

CRNN baseline result

SEND (Ours)

Our proposed model result

4

Demo4

Funk

Anchor

Original stereo mix

Reference

Human expert mix

DSP

DSP baseline result

CRNN

CRNN baseline result

SEND (Ours)

Our proposed model result

5

Demo5

Blues

Anchor

Original stereo mix

Reference

Human expert mix

DSP

DSP baseline result

CRNN

CRNN baseline result

SEND (Ours)

Our proposed model result

🎵 All music samples are royalty-free and used solely for research demonstration.