When testing with v0.2.1 version of the py_vnr model, I see an increase in the DUT vs REF inference output error. Earlier, the max error between dut and ref tflite prediction was around 0.03 and it's now around 0.05-0.06. I talked with @Allan-xmos and there doesn't seem to be a clear reason for this increase. As a result of this I've changed the pass criterions of a few of the VNR tests.
fixes https://github.com/xmos/fwk_voice/issues/336
When testing with v0.2.1 version of the py_vnr model, I see an increase in the DUT vs REF inference output error. Earlier, the max error between dut and ref tflite prediction was around 0.03 and it's now around 0.05-0.06. I talked with @Allan-xmos and there doesn't seem to be a clear reason for this increase. As a result of this I've changed the pass criterions of a few of the VNR tests.