Configurable Output#
Using MixSim can generate personalize data. Depending on the different tasks, the custom output can be achieved using the output_members.
Supported output members#
During the simulation, without affecting the simulation speed, MixSim will collect the supported output members as much as possible.
Based on the effectiveness, the number of output members can be stored in the disks, which is optimized by the arrow system.
| Method | Description |
|---|---|
mix_y |
signals mixed by dry clean source signals. |
mix_y_rvb |
signals mixed by reverberant clean signals. |
mix_y_rvb_direct_path |
like mix_y_rvb, but only with direct-path part. |
mix_y_rvb_early |
like mix_y_rvb, but only with early-path part. |
mix_y_rvb_late |
like mix_y_rvb, but only with late-path part. |
n_mix_y |
mix_y with noise added. |
n_mix_y_rvb |
mix_y_rvb with noise added. |
n_mix_y_rir_direct_path |
mix_y_rvb_direct_path with noise added. |
n_mix_y_rir_early |
mix_y_rvb_early with noise added. |
n_mix_y_rir_late |
mix_y_rvb_late with noise added. |
s_<id>_y |
dry clean signals with the id <id>. If the number of speakers is 1, you can use s_y. The same applies below |
s_<id>_y_rvb |
reverberant clean signals with the id <id>. |
s_<id>_y_rvb_direct_path |
like s_<id>_y_rvb, but only use direct-path part. |
s_<id>_y_rvb_early |
like s_<id>_y_rvb, but only use early reflection. |
s_<id>_y_rvb_late |
like s_<id>_y_rvb, but only use late reflection. |
s_<id>_rvb_peak_idx |
peak index of room impulse response of the source with the id <id>. |
s_<id>_y_vad |
voice activity detection result of the source with the id <id>. |
s_<id>_vad_label |
voice activity detection label (0 or 1) of the source with the id <id>. |
s_<id>_transcript |
transcript of the source with the id <id>. |
s_<id>_traj |
trajectory of the source with the id <id>. |
s_<id>_doa |
direction of arrival of the source with the id <id> (in degree). |
s_<id>_timestamps |
timestamps of the signal with the id <id>. |
n_y |
noise signals. |
n_gain |
noise gain. |
In your config file, you can specify the output_members.
For instance, if you want to do single-channel speech enhancement without reverberation, you may specify the following members:
from mixsim import ConfigurableWriter
output_members=["n_mix_y", "s_y", "n_y"]
For multichannel two-speaker localization/tracking with additive noise and reverberation, you may:
output_members=["n_mix_y_rvb", "s_1_traj", "s_2_traj", "s_1_doa", "s_2_doa", "n_y"]
For multichannel single-speaker speech recognition with noise and reverberation, you may:
output_members=["n_mix_y_rvb", "s_y", "s_transcript"]
For multichannel dereverberation with noise and reverberation, you may:
output_members=["n_mix_y_rvb", "s_y_rvb_direct_path"]
Then, you can initialize the configurable writer:
writer = ConfigurableWriter(output_members=output_members)