fanlu comments

Results 9 comments of


                                            fanlu

Add support for estimation of crp_alpha

How to fix the total number of speakers? In most call center scenario, there are only 2 or 3 speakers.

Utility for prefetching data on GPUs

@pzelasko Is there a way to store wavefoem in hdf5 format, then we can load data fast and do on-the-fly augmentation?

Test CTC/LF-MMI interpolation

I have met this same error on asg_train.py with aishell recipe training. ``` Traceback (most recent call last): File "asg_train.py", line 428, in main() File "asg_train.py", line 375, in main...

Test CTC/LF-MMI interpolation

This error occurred around 1w4 iterations in training procedure, the iterations is so large because I have changed the max_frames from 90000 to 10000 to avoid the OOM error. My...

use the latest code with Debug mode, ``` cmake -DCMAKE_BUILD_TYPE=Debug .. make -j ``` the problem below occured ``` nvlink error : Entry function '_ZN3cub31DeviceRadixSortSingleTileKernelINS_21DeviceRadixSortPolicyIiiiE9Policy700ELb0EiiiEEvPKT1_PS4_PKT2_PS8_T3_ii' uses too much shared data...

Test CTC/LF-MMI interpolation

I have build the latest code with Debug mode, and train asg_train.py with `K2_SYNC_KERNELS=1`, but the problem `RuntimeError: cuDNN error: CUDNN_STATUS_EXECUTION_FAILED` do not disappear. ``` export K2_SYNC_KERNELS=1;export CUDA_VISIBLE_DEVICES=2;PYTHONPATH=/mnt/cfs1_alias1/asr/users/fanlu/task/snowfall:$PYTHONPATH; python asg_train.py...