Data Preparation
Python Modules
Configuration Files
config.yaml
1path:
2 root: ???
3 metadata_dir: ${path.root}/clarity_data/metadata
4
5random_seed: 0
6shuffle_rooms: False
7
8# Build rooms
9room_datasets:
10 train:
11 rpf_location: ${path.root}/clarity_data/train/rooms/rpf
12 n_interferers: 3
13 n_rooms: 6000
14 start_room: 1
15 dev:
16 rpf_location: ${path.root}/clarity_data/dev/rooms/rpf
17 n_interferers: 3
18 n_rooms: 2500
19 start_room: 6001
20 demo:
21 rpf_location: ${path.root}/clarity_data/dev/rooms/rpf
22 n_interferers: 3
23 n_rooms: 200
24 start_room: 6001
25
26# Instantiate_scenes
27scene_datasets:
28 train:
29 n_scenes: 6000
30 room_selection: SEQUENTIAL
31 scene_start_index: 1
32 dev:
33 n_scenes: 2500
34 room_selection: SEQUENTIAL
35 scene_start_index: 6001
36 demo:
37 n_scenes: 100
38 room_selection: RANDOM
39 scene_start_index: 6001
40
41target:
42 target_speakers: ${path.metadata_dir}/target_speech_list.json
43 target_selection: SEQUENTIAL
44 pre_samples_range: [66150, 88200]
45 post_samples_range: [44100, 44100]
46
47snr_range:
48 train: [-12, 6]
49 dev: [-12, 6]
50 demo: [-12, 6]
51
52interferer:
53 speech_interferers: ${path.metadata_dir}/masker_speech_list.json
54 noise_interferers: ${path.metadata_dir}/masker_nonspeech_list.json
55 music_interferers: ${path.metadata_dir}/masker_music_list.json
56 number: [2, 3]
57 start_time_range: [0, 0]
58 end_early_time_range: [0, 0]
59
60listener:
61 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"]
62 channels: ["ED", "BTE_fr", "BTE_mid", "BTE_rear"]
63 # parameters all in units of samples or degrees
64 # The time at which the head turn starts relative to target speaker onset
65 relative_start_time_range: [-28004, 38147] # -0.635 s to 0.865 s
66 # Rotations have a normally distributed duration
67 duration_mean: 8820
68 duration_sd: 441 # 100 ms
69 # Head is initially pointing away from talker, uniform within a range
70 # Note it can be either offset to left or right - drawn at random
71 angle_initial_mean: 25
72 angle_initial_sd: 5
73 # Head turns to point at the speaker within some margin
74 angle_final_range: [0, 10]
75
76# Render scenes
77render_starting_chunk: 0 # there are (6000 train + 2500 dev) / (12 + 5) = 500 trunks in total. If multi_run, should be 0, 10, 20, 30, ..., 490 if render_n_chunk_to_process=10
78render_n_chunk_to_process: 10 # i.e. (12 train + 5 dev) * 10 scenes to render. If not multi_run, set 50
79
80scene_renderer:
81 train:
82 paths:
83 hoairs: ${path.root}/clarity_data/train/rooms/HOA_IRs
84 hrirs: ${path.root}/clarity_data/hrir/HRIRs_MAT
85 scenes: ${path.root}/clarity_data/train/scenes
86 targets: ${path.root}/clarity_data/train/targets
87 interferers: ${path.root}/clarity_data/train/interferers/{type}
88 metadata:
89 room_definitions: ${path.metadata_dir}/rooms.train.json
90 scene_definitions: ${path.metadata_dir}/scenes.train.json
91 hrir_metadata: ${path.metadata_dir}/hrir_data.json
92 chunk_size: 12
93 dev:
94 paths:
95 hoairs: ${path.root}/clarity_data/dev/rooms/HOA_IRs
96 hrirs: ${path.root}/clarity_data/hrir/HRIRs_MAT
97 scenes: ${path.root}/clarity_data/dev/scenes
98 targets: ${path.root}/clarity_data/dev/targets
99 interferers: ${path.root}/clarity_data/dev/interferers/{type}
100 metadata:
101 room_definitions: ${path.metadata_dir}/rooms.dev.json
102 scene_definitions: ${path.metadata_dir}/scenes.dev.json
103 hrir_metadata: ${path.metadata_dir}/hrir_data.json
104 chunk_size: 5
105 demo:
106 paths:
107 hoairs: ${path.root}/clarity_data/dev/rooms/HOA_IRs
108 hrirs: ${path.root}/clarity_data/hrir/HRIRs_MAT
109 scenes: ${path.root}/clarity_data/demo/scenes
110 targets: ${path.root}/clarity_data/dev/targets
111 interferers: ${path.root}/clarity_data/dev/interferers/{type}
112 metadata:
113 room_definitions: ${path.metadata_dir}/rooms.demo.json
114 scene_definitions: ${path.metadata_dir}/scenes.demo.json
115 hrir_metadata: ${path.metadata_dir}/hrir_data.json
116 chunk_size: 5
117
118render_params:
119 ambisonic_order: 6
120 equalise_loudness: False
121 reference_channel: 1
122 channel_norms: [12.0, 3.0, 3.0, 3.0]
123
124hydra:
125 run:
126 dir: .
127 job:
128 chdir: True
129
130defaults:
131 - override hydra/launcher: cec2_submitit_local
Sub-sections: