n times of 10s data inference, the result from the first 10 seconds will be different from the next ten seconds

breizhn / DTLN

Tensorflow 2.x implementation of the DTLN real time speech denoising model. With TF-lite, ONNX and real-time audio processing support.

MIT License

567 stars 160 forks source link

n times of 10s data inference, the result from the first 10 seconds will be different from the next ten seconds #59

Open yudashuixiao1 opened 2 years ago

yudashuixiao1 commented 2 years ago

Hi, n times of 10s data inference, the result from the first 10 seconds will be different from the next ten seconds. why？ how can I solve the problem?

![This is an image]F:\data\dns_dataset\测试\ceef8b398307404689d958aac0dcdf4e.png