Data Generation

Python Modules

recipes.cec2.baseline.data_generation.build_additional_scenes

recipes.cec2.baseline.data_generation.render_additional_scenes

Configuration Files

additional_data_config.yaml

 1path:
 2  root: ???
 3  metadata_dir: ${path.root}/clarity_data/metadata
 4  additional_data_file: ${path.metadata_dir}/scenes.train_additional.json
 5
 6random_seed: 0
 7shuffle_rooms: True
 8
 9# Instantiate_scenes
10scene_datasets:
11  train:
12    n_scenes: 10000
13    room_selection: RANDOM
14    scene_start_index: 10001
15
16target:
17  target_speakers: ${path.metadata_dir}/target_speech_list.json
18  target_selection: RANDOM
19  pre_samples_range: [66150, 88200]
20  post_samples_range: [44100, 44100]
21
22snr_range:
23  train: [-12, 6]
24
25interferer:
26  speech_interferers: ${path.metadata_dir}/masker_speech_list.json
27  noise_interferers: ${path.metadata_dir}/masker_nonspeech_list.json
28  music_interferers: ${path.metadata_dir}/masker_music_list.json
29  number: [2, 3]
30  start_time_range: [0, 0]
31  end_early_time_range: [0, 0]
32
33listener:
34  heads: ["BuK", "DADEC", "KEMAR", "VP_E1",  "VP_E2", "VP_E4", "VP_E5", "VP_E6", "VP_E7", "VP_E8", "VP_E9", "VP_E11", "VP_E12", "VP_E13", "VP_N1", "VP_N3", "VP_N4", "VP_N5", "VP_N6"]
35  channels: ["ED", "BTE_fr", "BTE_mid", "BTE_rear"]
36  # parameters all in units of samples or degrees
37  # The time at which the head turn starts relative to target speaker onset
38  relative_start_time_range: [-28004, 38147]  # -0.635 s to 0.865 s
39  # Rotations have a normally distributed duration
40  duration_mean: 8820
41  duration_sd: 441  # 100 ms
42  # Head is initially pointing away from talker, uniform within a range
43  # Note it can be either offset to left or right - drawn at random
44  angle_initial_mean: 25
45  angle_initial_sd: 5
46  # Head turns to point at the speaker within some margin
47  angle_final_range: [0, 10]
48
49
50# Render scenes
51render_starting_chunk: 0  # If multi_run, should be 0, 10, 20, 30, ..., 1000 if render_n_chunk_to_process=10 & 10000 scenes to render
52render_n_chunk_to_process: 200  # i.e.  (render_n_chunk_to_process * chunk_size) scenes to render
53
54scene_renderer:
55  train:
56    paths:
57      hoairs: ${path.root}/clarity_data/train/rooms/HOA_IRs
58      hrirs: ${path.root}/clarity_data/hrir/HRIRs_MAT
59      scenes: ${path.root}/clarity_data/train/additional_scenes
60      targets: ${path.root}/clarity_data/train/targets
61      interferers: ${path.root}/clarity_data/train/interferers/{type}
62    metadata:
63      room_definitions: ${path.metadata_dir}/rooms.train.json
64      scene_definitions: ${path.metadata_dir}/scenes.train_additional.json
65      hrir_metadata: ${path.metadata_dir}/hrir_data.json
66    chunk_size: 10
67
68render_params:
69  ambisonic_order: 6
70  equalise_loudness: False
71  reference_channel: 1
72  channel_norms: [12.0, 3.0, 3.0, 3.0]
73
74hydra:
75  run:
76    dir: .
77  job:
78    chdir: True
79
80defaults:
81  - override hydra/launcher: cec2_submitit_local

Data Generation Details: