justdata 1.0.0__py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (106) hide show
  1. justdata/__init__.py +3 -0
  2. justdata/acoustic/__init__.py +381 -0
  3. justdata/acoustic/adapters.py +329 -0
  4. justdata/acoustic/augment/__init__.py +119 -0
  5. justdata/acoustic/augment/batch.py +813 -0
  6. justdata/acoustic/augment/configs.py +204 -0
  7. justdata/acoustic/augment/patchout.py +331 -0
  8. justdata/acoustic/augment/spectrogram.py +806 -0
  9. justdata/acoustic/augment/waveform.py +879 -0
  10. justdata/acoustic/batching.py +27 -0
  11. justdata/acoustic/channel.py +53 -0
  12. justdata/acoustic/compat/__init__.py +95 -0
  13. justdata/acoustic/compat/ast.py +626 -0
  14. justdata/acoustic/compat/ced.py +152 -0
  15. justdata/acoustic/compat/efficientat.py +243 -0
  16. justdata/acoustic/compat/panns.py +115 -0
  17. justdata/acoustic/compat/passt.py +191 -0
  18. justdata/acoustic/configs.py +519 -0
  19. justdata/acoustic/corruptions/__init__.py +33 -0
  20. justdata/acoustic/corruptions/codec.py +92 -0
  21. justdata/acoustic/corruptions/datasets.py +155 -0
  22. justdata/acoustic/corruptions/device.py +39 -0
  23. justdata/acoustic/corruptions/dynamics.py +55 -0
  24. justdata/acoustic/corruptions/filters.py +105 -0
  25. justdata/acoustic/corruptions/noise.py +173 -0
  26. justdata/acoustic/corruptions/registry.py +230 -0
  27. justdata/acoustic/corruptions/reverb.py +49 -0
  28. justdata/acoustic/corruptions/timefreq.py +127 -0
  29. justdata/acoustic/datasets.py +52 -0
  30. justdata/acoustic/dcase2025.py +814 -0
  31. justdata/acoustic/decoding.py +73 -0
  32. justdata/acoustic/eval_views.py +54 -0
  33. justdata/acoustic/evals.py +71 -0
  34. justdata/acoustic/filters.py +18 -0
  35. justdata/acoustic/frontends/__init__.py +37 -0
  36. justdata/acoustic/frontends/ast.py +116 -0
  37. justdata/acoustic/frontends/kaldi.py +34 -0
  38. justdata/acoustic/frontends/log.py +94 -0
  39. justdata/acoustic/frontends/mel.py +173 -0
  40. justdata/acoustic/frontends/mfcc.py +26 -0
  41. justdata/acoustic/frontends/pcen.py +26 -0
  42. justdata/acoustic/frontends/stft.py +174 -0
  43. justdata/acoustic/jax.py +23 -0
  44. justdata/acoustic/labels.py +169 -0
  45. justdata/acoustic/layouts.py +86 -0
  46. justdata/acoustic/metadata.py +152 -0
  47. justdata/acoustic/normalization.py +196 -0
  48. justdata/acoustic/pipelines.py +112 -0
  49. justdata/acoustic/postprocessing.py +328 -0
  50. justdata/acoustic/preprocessing.py +175 -0
  51. justdata/acoustic/presets.py +257 -0
  52. justdata/acoustic/registry.py +381 -0
  53. justdata/acoustic/resampling.py +218 -0
  54. justdata/acoustic/schema.py +51 -0
  55. justdata/acoustic/segment.py +369 -0
  56. justdata/acoustic/sources.py +503 -0
  57. justdata/acoustic/stages.py +59 -0
  58. justdata/acoustic/stats.py +50 -0
  59. justdata/acoustic/tasks.py +375 -0
  60. justdata/audio/__init__.py +20 -0
  61. justdata/core/__init__.py +74 -0
  62. justdata/core/adapters.py +46 -0
  63. justdata/core/augmentations.py +34 -0
  64. justdata/core/filters.py +94 -0
  65. justdata/core/finalization.py +173 -0
  66. justdata/core/label_mixing.py +129 -0
  67. justdata/core/loader.py +384 -0
  68. justdata/core/metadata.py +250 -0
  69. justdata/core/presets.py +155 -0
  70. justdata/core/registry.py +229 -0
  71. justdata/core/sources.py +87 -0
  72. justdata/core/stats.py +227 -0
  73. justdata/corruptions.py +52 -0
  74. justdata/py.typed +0 -0
  75. justdata/vision/__init__.py +17 -0
  76. justdata/vision/augmentations/__init__.py +31 -0
  77. justdata/vision/augmentations/auto.py +480 -0
  78. justdata/vision/augmentations/color.py +163 -0
  79. justdata/vision/augmentations/composed.py +156 -0
  80. justdata/vision/augmentations/geometric.py +302 -0
  81. justdata/vision/augmentations/mixing.py +232 -0
  82. justdata/vision/augmentations/registry.py +142 -0
  83. justdata/vision/corruptions/__init__.py +16 -0
  84. justdata/vision/corruptions/blur.py +65 -0
  85. justdata/vision/corruptions/digital.py +26 -0
  86. justdata/vision/corruptions/noise.py +20 -0
  87. justdata/vision/corruptions/registry.py +113 -0
  88. justdata/vision/corruptions/weather.py +100 -0
  89. justdata/vision/datasets.py +34 -0
  90. justdata/vision/encodings/__init__.py +1 -0
  91. justdata/vision/encodings/heatmaps.py +196 -0
  92. justdata/vision/minic.py +69 -0
  93. justdata/vision/pipelines.py +62 -0
  94. justdata/vision/presets.py +509 -0
  95. justdata/vision/sources.py +1080 -0
  96. justdata/vision/stages.py +271 -0
  97. justdata/vision/tasks/__init__.py +1 -0
  98. justdata/vision/tasks/classification.py +321 -0
  99. justdata/vision/tasks/depth.py +25 -0
  100. justdata/vision/tasks/segmentation.py +194 -0
  101. justdata/vision/transforms.py +113 -0
  102. justdata/vision/utils.py +741 -0
  103. justdata-1.0.0.dist-info/METADATA +860 -0
  104. justdata-1.0.0.dist-info/RECORD +106 -0
  105. justdata-1.0.0.dist-info/WHEEL +4 -0
  106. justdata-1.0.0.dist-info/licenses/LICENSE +23 -0
@@ -0,0 +1,155 @@
1
+ from __future__ import annotations
2
+
3
+ from typing import Literal, Union
4
+
5
+ import tensorflow as tf
6
+
7
+ from justdata.acoustic.corruptions.registry import (
8
+ _metadata_with_corruption,
9
+ _seed_from_index,
10
+ apply_audio_corruption,
11
+ )
12
+ from justdata.acoustic.schema import FEATURES, SAMPLE_RATE, WAVEFORM
13
+ from justdata.core.loader import load_ds
14
+ from justdata.core.registry import get_pipeline
15
+
16
+
17
+ def create_audio_corruption_datasets(
18
+ corruption_types: Union[str, list[str]],
19
+ severity: int,
20
+ base_dataset: str,
21
+ preset: str | None,
22
+ *,
23
+ split: str = "validation",
24
+ seed: int = 0,
25
+ corruption_domain: Literal["waveform", "spectrogram"] = "waveform",
26
+ **load_kwargs,
27
+ ) -> Union[tuple[tf.data.Dataset, int], tuple[list[tf.data.Dataset], int]]:
28
+ """Create severity-parameterized acoustic corruption evaluation datasets."""
29
+ if corruption_domain not in {"waveform", "spectrogram"}:
30
+ raise ValueError("corruption_domain must be 'waveform' or 'spectrogram'.")
31
+
32
+ kwargs = dict(load_kwargs)
33
+ corruption_configs = kwargs.pop("corruption_configs", None)
34
+ kwargs["return_raw_ds"] = True
35
+ kwargs.setdefault("dataset_names_arg", base_dataset)
36
+ kwargs.setdefault("splits_arg", split)
37
+ kwargs.setdefault("dataset_type", "train" if split == "train" else "validation")
38
+ kwargs.setdefault("batch_size", 32)
39
+ kwargs.setdefault("seed", seed)
40
+ kwargs.setdefault("deterministic", True)
41
+
42
+ if "pipeline" not in kwargs and "postprocess_fn" not in kwargs:
43
+ kwargs["pipeline"] = get_pipeline(dataset=base_dataset, preset=preset)
44
+
45
+ ds, tools = load_ds(**kwargs)
46
+ finalize_fn = tools["finalize_fn"]
47
+
48
+ return_list = isinstance(corruption_types, list)
49
+ names = corruption_types if return_list else [corruption_types]
50
+ datasets_out = []
51
+
52
+ for index, c_name in enumerate(names):
53
+ config = None
54
+ if isinstance(corruption_configs, dict):
55
+ config = corruption_configs.get(c_name)
56
+
57
+ if corruption_domain == "waveform":
58
+ ds_c = ds.enumerate().map(
59
+ lambda i, sample, c_name=c_name, config=config, index=index: (
60
+ _corrupt_sample(
61
+ i,
62
+ sample,
63
+ c_name,
64
+ severity,
65
+ seed,
66
+ "waveform",
67
+ WAVEFORM,
68
+ index,
69
+ config,
70
+ )
71
+ ),
72
+ num_parallel_calls=tf.data.AUTOTUNE,
73
+ deterministic=True,
74
+ )
75
+ ds_c, n_batches = finalize_fn(ds_c)
76
+ else:
77
+
78
+ def corrupt_after_postprocess(
79
+ dataset,
80
+ c_name=c_name,
81
+ config=config,
82
+ index=index,
83
+ ):
84
+ return dataset.enumerate().map(
85
+ lambda i, sample: _corrupt_sample(
86
+ i,
87
+ sample,
88
+ c_name,
89
+ severity,
90
+ seed,
91
+ "spectrogram",
92
+ FEATURES,
93
+ index,
94
+ config,
95
+ ),
96
+ num_parallel_calls=tf.data.AUTOTUNE,
97
+ deterministic=True,
98
+ )
99
+
100
+ ds_c, n_batches = finalize_fn(
101
+ ds,
102
+ post_postprocess_transform=corrupt_after_postprocess,
103
+ )
104
+
105
+ datasets_out.append(ds_c)
106
+
107
+ if not datasets_out:
108
+ n_batches = 0
109
+ if return_list:
110
+ return datasets_out, n_batches
111
+ return datasets_out[0], n_batches
112
+
113
+
114
+ def _corrupt_sample(
115
+ index: tf.Tensor,
116
+ sample: dict,
117
+ name: str,
118
+ severity: int,
119
+ seed: int,
120
+ domain: str,
121
+ key: str,
122
+ salt: int,
123
+ config: dict | None,
124
+ ) -> dict:
125
+ if key not in sample:
126
+ raise ValueError(
127
+ f"Audio corruption domain '{domain}' requires sample key '{key}'."
128
+ )
129
+ corruption_config = config
130
+ if domain == "waveform":
131
+ if SAMPLE_RATE not in sample:
132
+ raise ValueError(
133
+ f"Audio corruption domain '{domain}' requires sample key "
134
+ f"'{SAMPLE_RATE}'."
135
+ )
136
+ corruption_config = dict(config) if config is not None else {}
137
+ corruption_config.setdefault("sample_rate", sample[SAMPLE_RATE])
138
+ sample_seed = _seed_from_index(seed, index, salt=salt * 1_000_003)
139
+ result = dict(sample)
140
+ result[key] = apply_audio_corruption(
141
+ result[key],
142
+ name,
143
+ severity,
144
+ sample_seed,
145
+ config=corruption_config,
146
+ )
147
+ return _metadata_with_corruption(
148
+ result,
149
+ name=name,
150
+ severity=severity,
151
+ domain=domain,
152
+ )
153
+
154
+
155
+ __all__ = ["create_audio_corruption_datasets"]
@@ -0,0 +1,39 @@
1
+ from __future__ import annotations
2
+
3
+ import tensorflow as tf
4
+
5
+ from justdata.acoustic.corruptions.registry import (
6
+ _convolve_waveform,
7
+ _severity_value,
8
+ register_audio_corruption,
9
+ )
10
+
11
+
12
+ DEVICE_IR_SET = {
13
+ 1: (0.02, 0.08, 0.80, 0.08, 0.02),
14
+ 2: (0.04, 0.12, 0.68, 0.12, 0.04),
15
+ 3: (0.08, 0.16, 0.52, 0.16, 0.08),
16
+ 4: (0.12, 0.20, 0.36, 0.20, 0.12),
17
+ 5: (0.16, 0.22, 0.24, 0.22, 0.16),
18
+ }
19
+
20
+
21
+ @register_audio_corruption("device_ir")
22
+ def device_ir(
23
+ audio_or_features: tf.Tensor,
24
+ severity: int,
25
+ seed: tf.Tensor,
26
+ config: dict | None = None,
27
+ ) -> tf.Tensor:
28
+ del seed
29
+ if config is not None and "ir" in config:
30
+ ir = tf.reshape(tf.cast(config["ir"], tf.float32), [-1])
31
+ else:
32
+ ir = _severity_value(DEVICE_IR_SET, severity)
33
+ return _convolve_waveform(audio_or_features, ir, compensate_delay=False)
34
+
35
+
36
+ __all__ = [
37
+ "DEVICE_IR_SET",
38
+ "device_ir",
39
+ ]
@@ -0,0 +1,55 @@
1
+ from __future__ import annotations
2
+
3
+ import tensorflow as tf
4
+
5
+ from justdata.acoustic.corruptions.registry import (
6
+ _EPS,
7
+ _severity_value,
8
+ register_audio_corruption,
9
+ )
10
+
11
+
12
+ CLIPPING_THRESHOLD = {1: 0.95, 2: 0.8, 3: 0.6, 4: 0.4, 5: 0.25}
13
+ DRC_THRESHOLD_DB = {1: -3, 2: -6, 3: -12, 4: -18, 5: -24}
14
+ DRC_RATIO = {1: 1.5, 2: 2.0, 3: 4.0, 4: 8.0, 5: 12.0}
15
+
16
+
17
+ @register_audio_corruption("clipping")
18
+ def clipping(
19
+ audio_or_features: tf.Tensor,
20
+ severity: int,
21
+ seed: tf.Tensor,
22
+ config: dict | None = None,
23
+ ) -> tf.Tensor:
24
+ del seed, config
25
+ threshold = _severity_value(CLIPPING_THRESHOLD, severity)
26
+ x = tf.cast(audio_or_features, tf.float32)
27
+ return tf.clip_by_value(x, -threshold, threshold)
28
+
29
+
30
+ @register_audio_corruption("dynamic_range_compression")
31
+ def dynamic_range_compression(
32
+ audio_or_features: tf.Tensor,
33
+ severity: int,
34
+ seed: tf.Tensor,
35
+ config: dict | None = None,
36
+ ) -> tf.Tensor:
37
+ del seed, config
38
+ threshold_db = _severity_value(DRC_THRESHOLD_DB, severity)
39
+ ratio = _severity_value(DRC_RATIO, severity)
40
+ x = tf.cast(audio_or_features, tf.float32)
41
+ magnitude = tf.abs(x)
42
+ level_db = 20.0 * tf.math.log(magnitude + _EPS) / tf.math.log(10.0)
43
+ over = level_db - threshold_db
44
+ gain_db = tf.where(over > 0.0, over / ratio - over, tf.zeros_like(over))
45
+ gain = tf.pow(tf.constant(10.0, dtype=tf.float32), gain_db / 20.0)
46
+ return tf.sign(x) * magnitude * gain
47
+
48
+
49
+ __all__ = [
50
+ "CLIPPING_THRESHOLD",
51
+ "DRC_RATIO",
52
+ "DRC_THRESHOLD_DB",
53
+ "clipping",
54
+ "dynamic_range_compression",
55
+ ]
@@ -0,0 +1,105 @@
1
+ from __future__ import annotations
2
+
3
+ import tensorflow as tf
4
+
5
+ from justdata.acoustic.corruptions.registry import (
6
+ _frequency_bins,
7
+ _rfft_filter,
8
+ _sample_rate,
9
+ _severity_value,
10
+ register_audio_corruption,
11
+ )
12
+
13
+
14
+ BANDPASS_HZ_32K = {
15
+ 1: (80, 14000),
16
+ 2: (120, 10000),
17
+ 3: (200, 7000),
18
+ 4: (400, 5000),
19
+ 5: (800, 3500),
20
+ }
21
+ LOWPASS_HZ_32K = {1: 12000, 2: 9000, 3: 6000, 4: 4000, 5: 2500}
22
+ HIGHPASS_HZ_32K = {1: 40, 2: 120, 3: 400, 4: 1000, 5: 2500}
23
+ EQUALIZATION_TILT_DB_PER_OCTAVE = {1: -1, 2: -2, 3: -4, 4: -6, 5: -9}
24
+
25
+
26
+ @register_audio_corruption("low_pass")
27
+ def low_pass(
28
+ audio_or_features: tf.Tensor,
29
+ severity: int,
30
+ seed: tf.Tensor,
31
+ config: dict | None = None,
32
+ ) -> tf.Tensor:
33
+ del seed
34
+ sample_rate = _sample_rate(config)
35
+ cutoff = _severity_value(LOWPASS_HZ_32K, severity)
36
+ freqs = _frequency_bins(audio_or_features, sample_rate)
37
+ response = tf.cast(freqs <= cutoff, tf.float32)
38
+ return _rfft_filter(audio_or_features, response)
39
+
40
+
41
+ @register_audio_corruption("high_pass")
42
+ def high_pass(
43
+ audio_or_features: tf.Tensor,
44
+ severity: int,
45
+ seed: tf.Tensor,
46
+ config: dict | None = None,
47
+ ) -> tf.Tensor:
48
+ del seed
49
+ sample_rate = _sample_rate(config)
50
+ cutoff = _severity_value(HIGHPASS_HZ_32K, severity)
51
+ freqs = _frequency_bins(audio_or_features, sample_rate)
52
+ response = tf.cast(freqs >= cutoff, tf.float32)
53
+ return _rfft_filter(audio_or_features, response)
54
+
55
+
56
+ @register_audio_corruption("band_pass")
57
+ def band_pass(
58
+ audio_or_features: tf.Tensor,
59
+ severity: int,
60
+ seed: tf.Tensor,
61
+ config: dict | None = None,
62
+ ) -> tf.Tensor:
63
+ del seed
64
+ sample_rate = _sample_rate(config)
65
+ low_high = _severity_value(BANDPASS_HZ_32K, severity)
66
+ low = low_high[0]
67
+ high = low_high[1]
68
+ freqs = _frequency_bins(audio_or_features, sample_rate)
69
+ response = tf.cast(tf.logical_and(freqs >= low, freqs <= high), tf.float32)
70
+ return _rfft_filter(audio_or_features, response)
71
+
72
+
73
+ @register_audio_corruption("equalization_tilt")
74
+ def equalization_tilt(
75
+ audio_or_features: tf.Tensor,
76
+ severity: int,
77
+ seed: tf.Tensor,
78
+ config: dict | None = None,
79
+ ) -> tf.Tensor:
80
+ del seed
81
+ sample_rate = _sample_rate(config)
82
+ ref_hz = 1000.0 if config is None else float(config.get("reference_hz", 1000.0))
83
+ tilt = _severity_value(EQUALIZATION_TILT_DB_PER_OCTAVE, severity)
84
+ freqs = _frequency_bins(audio_or_features, sample_rate)
85
+ safe_freqs = tf.maximum(freqs, tf.constant(20.0, dtype=tf.float32))
86
+ octaves = tf.math.log(safe_freqs / tf.cast(ref_hz, tf.float32)) / tf.math.log(2.0)
87
+ gain_db = tilt * octaves
88
+ response = tf.clip_by_value(
89
+ tf.pow(tf.constant(10.0, dtype=tf.float32), gain_db / 20.0),
90
+ 0.125,
91
+ 4.0,
92
+ )
93
+ return _rfft_filter(audio_or_features, response)
94
+
95
+
96
+ __all__ = [
97
+ "BANDPASS_HZ_32K",
98
+ "EQUALIZATION_TILT_DB_PER_OCTAVE",
99
+ "HIGHPASS_HZ_32K",
100
+ "LOWPASS_HZ_32K",
101
+ "band_pass",
102
+ "equalization_tilt",
103
+ "high_pass",
104
+ "low_pass",
105
+ ]
@@ -0,0 +1,173 @@
1
+ from __future__ import annotations
2
+
3
+ import tensorflow as tf
4
+
5
+ from justdata.acoustic.corruptions.registry import (
6
+ _EPS,
7
+ _as_tc,
8
+ _restore_rank,
9
+ _severity_value,
10
+ _split_seed,
11
+ register_audio_corruption,
12
+ )
13
+
14
+
15
+ WHITE_NOISE_SNR_DB = {1: 30, 2: 20, 3: 10, 4: 5, 5: 0}
16
+ PINK_NOISE_SNR_DB = {1: 30, 2: 20, 3: 10, 4: 5, 5: 0}
17
+ BROWN_NOISE_SNR_DB = {1: 30, 2: 20, 3: 10, 4: 5, 5: 0}
18
+ BACKGROUND_NOISE_SNR_DB = {1: 25, 2: 18, 3: 12, 4: 6, 5: 0}
19
+
20
+
21
+ def _normalize_noise(noise: tf.Tensor) -> tf.Tensor:
22
+ noise = tf.cast(noise, tf.float32)
23
+ axes = tf.range(tf.rank(noise))
24
+ noise = noise - tf.reduce_mean(noise, axis=axes, keepdims=True)
25
+ rms = tf.sqrt(tf.reduce_mean(tf.square(noise), axis=axes, keepdims=True))
26
+ return tf.math.divide_no_nan(noise, tf.maximum(rms, _EPS))
27
+
28
+
29
+ def _colored_noise(shape: tf.Tensor, seed: tf.Tensor, kind: str) -> tf.Tensor:
30
+ noise = tf.random.stateless_normal(shape, seed=seed, dtype=tf.float32)
31
+ if kind == "white":
32
+ return _normalize_noise(noise)
33
+
34
+ audio, rank = _as_tc(noise)
35
+ if kind == "brown":
36
+ return _restore_rank(_normalize_noise(tf.cumsum(audio, axis=0)), rank)
37
+ if kind != "pink":
38
+ raise ValueError("noise kind must be one of 'white', 'pink', or 'brown'.")
39
+
40
+ time = tf.shape(audio)[0]
41
+ channels_first = tf.transpose(audio, [1, 0])
42
+ spectrum = tf.signal.rfft(channels_first, fft_length=[time])
43
+ num_bins = tf.shape(spectrum)[-1]
44
+ freqs = tf.cast(tf.range(num_bins), tf.float32)
45
+ weights = tf.where(freqs > 0.0, tf.math.rsqrt(freqs), tf.zeros_like(freqs))
46
+ pink = tf.signal.irfft(
47
+ spectrum * tf.cast(weights[tf.newaxis, :], spectrum.dtype),
48
+ fft_length=[time],
49
+ )
50
+ return _restore_rank(_normalize_noise(tf.transpose(pink, [1, 0])), rank)
51
+
52
+
53
+ def _add_at_snr(
54
+ audio_or_features: tf.Tensor, noise: tf.Tensor, snr_db: tf.Tensor
55
+ ) -> tf.Tensor:
56
+ x = tf.cast(audio_or_features, tf.float32)
57
+ noise = tf.cast(noise, tf.float32)
58
+ signal_power = tf.reduce_mean(tf.square(x))
59
+ noise_power = tf.reduce_mean(tf.square(noise))
60
+ target_ratio = tf.pow(tf.constant(10.0, dtype=tf.float32), snr_db / 10.0)
61
+ scale = tf.sqrt(tf.math.divide_no_nan(signal_power, noise_power * target_ratio))
62
+ return x + noise * scale
63
+
64
+
65
+ def _add_colored_noise(
66
+ audio_or_features: tf.Tensor,
67
+ severity: int,
68
+ seed: tf.Tensor,
69
+ *,
70
+ kind: str,
71
+ snr_table: dict[int, int],
72
+ ) -> tf.Tensor:
73
+ snr_db = _severity_value(snr_table, severity)
74
+ noise = _colored_noise(tf.shape(audio_or_features), seed, kind)
75
+ return _add_at_snr(audio_or_features, noise, snr_db)
76
+
77
+
78
+ @register_audio_corruption("additive_white_noise")
79
+ def additive_white_noise(
80
+ audio_or_features: tf.Tensor,
81
+ severity: int,
82
+ seed: tf.Tensor,
83
+ config: dict | None = None,
84
+ ) -> tf.Tensor:
85
+ del config
86
+ return _add_colored_noise(
87
+ audio_or_features,
88
+ severity,
89
+ seed,
90
+ kind="white",
91
+ snr_table=WHITE_NOISE_SNR_DB,
92
+ )
93
+
94
+
95
+ @register_audio_corruption("additive_pink_noise")
96
+ def additive_pink_noise(
97
+ audio_or_features: tf.Tensor,
98
+ severity: int,
99
+ seed: tf.Tensor,
100
+ config: dict | None = None,
101
+ ) -> tf.Tensor:
102
+ del config
103
+ return _add_colored_noise(
104
+ audio_or_features,
105
+ severity,
106
+ seed,
107
+ kind="pink",
108
+ snr_table=PINK_NOISE_SNR_DB,
109
+ )
110
+
111
+
112
+ @register_audio_corruption("additive_brown_noise")
113
+ def additive_brown_noise(
114
+ audio_or_features: tf.Tensor,
115
+ severity: int,
116
+ seed: tf.Tensor,
117
+ config: dict | None = None,
118
+ ) -> tf.Tensor:
119
+ del config
120
+ return _add_colored_noise(
121
+ audio_or_features,
122
+ severity,
123
+ seed,
124
+ kind="brown",
125
+ snr_table=BROWN_NOISE_SNR_DB,
126
+ )
127
+
128
+
129
+ @register_audio_corruption("background_noise")
130
+ def background_noise(
131
+ audio_or_features: tf.Tensor,
132
+ severity: int,
133
+ seed: tf.Tensor,
134
+ config: dict | None = None,
135
+ ) -> tf.Tensor:
136
+ snr_db = _severity_value(BACKGROUND_NOISE_SNR_DB, severity)
137
+ noise_seed, hum_seed = tf.unstack(_split_seed(seed, 2))
138
+
139
+ configured = None if config is None else config.get("background")
140
+ if configured is not None:
141
+ background = tf.cast(configured, tf.float32)
142
+ else:
143
+ x = tf.convert_to_tensor(audio_or_features)
144
+ background = _colored_noise(tf.shape(x), noise_seed, "pink")
145
+ time = tf.cast(tf.shape(x)[0], tf.float32)
146
+ positions = tf.range(tf.shape(x)[0], dtype=tf.float32)
147
+ cycles = tf.random.stateless_uniform(
148
+ [],
149
+ seed=hum_seed,
150
+ minval=1.0,
151
+ maxval=8.0,
152
+ dtype=tf.float32,
153
+ )
154
+ hum = tf.sin(
155
+ 2.0 * 3.141592653589793 * cycles * positions / tf.maximum(time, 1.0)
156
+ )
157
+ if x.shape.rank == 2:
158
+ hum = hum[..., tf.newaxis]
159
+ background = background + 0.25 * hum
160
+
161
+ return _add_at_snr(audio_or_features, background, snr_db)
162
+
163
+
164
+ __all__ = [
165
+ "BACKGROUND_NOISE_SNR_DB",
166
+ "BROWN_NOISE_SNR_DB",
167
+ "PINK_NOISE_SNR_DB",
168
+ "WHITE_NOISE_SNR_DB",
169
+ "additive_brown_noise",
170
+ "additive_pink_noise",
171
+ "additive_white_noise",
172
+ "background_noise",
173
+ ]