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
585 stars 161 forks source link

about problem of adding win function frame #78

Open yudashuixiao1 opened 6 months ago

yudashuixiao1 commented 6 months ago

Why doesn't this model need a window function? The original model is trained without any modifications, but the inference model introduces weak background noise. What could be the reason for this? The inference performance of the model trained with a window function in the .h5 format is fine, but when converted to a tflite model with a window function, issues arise.