xiph
How to convert the model to rnnoise_data_little.c
Firstly, thank you for updating the rnnoise, Now I can convert the model to rnnoise_data.c by using dump_rnnoise_weights.py, I want to konw how to convert the model to rnnoise_data_little.c
It's the same conversion. It's just that the "little" models were trained with sparsification (which isn't yet well documented, sorry)
@BancoLin thank you for your advice, I have another question, in my train script the sparsify_start ,sparsify_stop , sparsify_interval, and sparsify_exponent remain unchanged, while the epochs are 200 and each epoch takes 12500 steps, when i dump the checkpoint file rnnoise_2.pth into the rnnoise_data_little.c, the array length is not the same as writer provided, for example: the length of gru1_input_weights_float i got is 147456,
static const float gru1_input_weights_float[147456]
while in write's file, is:
static const float gru1_input_weights_float[58976]
how can i reduce the length?
@BancoLin thank you for your advice, I have another question, in my train script the
sparsify_start,sparsify_stop,sparsify_interval, andsparsify_exponentremain unchanged, while the epochs are200and each epoch takes12500steps, when i dump the checkpoint filernnoise_2.pthinto thernnoise_data_little.c, the array length is not the same as writer provided, for example: the length ofgru1_input_weights_floati got is 147456,static const float gru1_input_weights_float[147456]
while in write's file, is:
static const float gru1_input_weights_float[58976]
how can i reduce the length?
Did you dump the last checkpoint file to check the length?
Dumping rnnoise_2.pth seems too early, the model hasn't been sparsified yet.
The number of weights in gru1_input_weights_float should be equal to 3densitynb_inputs*nb_outputs.