sonusai 0.18.9__py3-none-any.whl → 0.19.6__py3-none-any.whl

sonusai/mixture/generation.py

@@ -1,24 +1,25 @@
-from sonusai.mixture.datatypes import AudioT
-from sonusai.mixture.datatypes import AudiosT
-from sonusai.mixture.datatypes import Augmentation
-from sonusai.mixture.datatypes import AugmentationRules
-from sonusai.mixture.datatypes import AugmentedTargets
-from sonusai.mixture.datatypes import GenMixData
-from sonusai.mixture.datatypes import ImpulseResponseFiles
-from sonusai.mixture.datatypes import Mixture
-from sonusai.mixture.datatypes import Mixtures
-from sonusai.mixture.datatypes import NoiseFiles
-from sonusai.mixture.datatypes import SpectralMasks
-from sonusai.mixture.datatypes import TargetFiles
-from sonusai.mixture.datatypes import Targets
-from sonusai.mixture.datatypes import UniversalSNRGenerator
-from sonusai.mixture.mixdb import MixtureDatabase
+# ruff: noqa: S608
+from .datatypes import AudiosT
+from .datatypes import AudioT
+from .datatypes import Augmentation
+from .datatypes import AugmentationRules
+from .datatypes import AugmentedTargets
+from .datatypes import GenMixData
+from .datatypes import ImpulseResponseFiles
+from .datatypes import Mixture
+from .datatypes import Mixtures
+from .datatypes import NoiseFiles
+from .datatypes import SpectralMasks
+from .datatypes import TargetFiles
+from .datatypes import Targets
+from .datatypes import UniversalSNRGenerator
+from .mixdb import MixtureDatabase
 
 
 def config_file(location: str) -> str:
     from os.path import join
 
-    return join(location, 'config.yml')
+    return join(location, "config.yml")
 
 
 def initialize_db(location: str, test: bool = False) -> None:
@@ -27,9 +28,16 @@ def initialize_db(location: str, test: bool = False) -> None:
     con = db_connection(location=location, create=True, test=test)
 
     con.execute("""
-    CREATE TABLE truth_setting(
+    CREATE TABLE truth_config(
     id INTEGER PRIMARY KEY NOT NULL,
-    setting TEXT NOT NULL)
+    config TEXT NOT NULL)
+    """)
+
+    con.execute("""
+    CREATE TABLE truth_parameters(
+    id INTEGER PRIMARY KEY NOT NULL,
+    name TEXT NOT NULL,
+    parameters INTEGER NOT NULL)
     """)
 
     con.execute("""
@@ -37,6 +45,7 @@ def initialize_db(location: str, test: bool = False) -> None:
     id INTEGER PRIMARY KEY NOT NULL,
     name TEXT NOT NULL,
     samples INTEGER NOT NULL,
+    class_indices TEXT NOT NULL,
     level_type TEXT NOT NULL,
     speaker_id INTEGER,
     FOREIGN KEY(speaker_id) REFERENCES speaker (id))
@@ -65,8 +74,6 @@ def initialize_db(location: str, test: bool = False) -> None:
     noise_mix_mode TEXT NOT NULL,
     num_classes INTEGER NOT NULL,
     seed INTEGER NOT NULL,
-    truth_mutex BOOLEAN NOT NULL,
-    truth_reduction_function TEXT NOT NULL,
     mixid_width INTEGER NOT NULL,
     speaker_metadata_tiers TEXT NOT NULL,
     textgrid_metadata_tiers TEXT NOT NULL)
@@ -87,7 +94,8 @@ def initialize_db(location: str, test: bool = False) -> None:
     con.execute("""
     CREATE TABLE impulse_response_file (
     id INTEGER PRIMARY KEY NOT NULL,
-    file TEXT NOT NULL)
+    file TEXT NOT NULL,
+    tags TEXT NOT NULL)
     """)
 
     con.execute("""
@@ -101,11 +109,11 @@ def initialize_db(location: str, test: bool = False) -> None:
     """)
 
     con.execute("""
-    CREATE TABLE target_file_truth_setting (
+    CREATE TABLE target_file_truth_config (
     target_file_id INTEGER,
-    truth_setting_id INTEGER,
+    truth_config_id INTEGER,
     FOREIGN KEY(target_file_id) REFERENCES target_file (id),
-    FOREIGN KEY(truth_setting_id) REFERENCES truth_setting (id))
+    FOREIGN KEY(truth_config_id) REFERENCES truth_config (id))
     """)
 
     con.execute("""
@@ -148,59 +156,55 @@ def initialize_db(location: str, test: bool = False) -> None:
 
 
 def populate_top_table(location: str, config: dict, test: bool = False) -> None:
-    """Populate top table
-    """
+    """Populate top table"""
     import json
 
-    from sonusai import SonusAIError
+    from .constants import MIXDB_VERSION
     from .mixdb import db_connection
 
-    if config['truth_mode'] not in ['normal', 'mutex']:
-        raise SonusAIError(f'invalid truth_mode: {config["truth_mode"]}')
-    truth_mutex = config['truth_mode'] == 'mutex'
-
     con = db_connection(location=location, readonly=False, test=test)
-    con.execute("""
+    con.execute(
+        """
     INSERT INTO top (version, asr_configs, class_balancing, feature, noise_mix_mode, num_classes,
-    seed, truth_mutex, truth_reduction_function, mixid_width, speaker_metadata_tiers, textgrid_metadata_tiers)
-    VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
-    """, (
-        1,
-        json.dumps(config['asr_configs']),
-        config['class_balancing'],
-        config['feature'],
-        config['noise_mix_mode'],
-        config['num_classes'],
-        config['seed'],
-        truth_mutex,
-        config['truth_reduction_function'],
-        0,
-        '',
-        ''))
+    seed, mixid_width, speaker_metadata_tiers, textgrid_metadata_tiers)
+    VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+    """,
+        (
+            MIXDB_VERSION,
+            json.dumps(config["asr_configs"]),
+            config["class_balancing"],
+            config["feature"],
+            config["noise_mix_mode"],
+            config["num_classes"],
+            config["seed"],
+            0,
+            "",
+            "",
+        ),
+    )
     con.commit()
     con.close()
 
 
 def populate_class_label_table(location: str, config: dict, test: bool = False) -> None:
-    """Populate class_label table
-    """
+    """Populate class_label table"""
     from .mixdb import db_connection
 
     con = db_connection(location=location, readonly=False, test=test)
-    con.executemany("INSERT INTO class_label (label) VALUES (?)",
-                    [(item,) for item in config['class_labels']])
+    con.executemany(
+        "INSERT INTO class_label (label) VALUES (?)",
+        [(item,) for item in config["class_labels"]],
+    )
     con.commit()
     con.close()
 
 
 def populate_class_weights_threshold_table(location: str, config: dict, test: bool = False) -> None:
-    """Populate class_weights_threshold table
-    """
-    from sonusai import SonusAIError
+    """Populate class_weights_threshold table"""
     from .mixdb import db_connection
 
-    class_weights_threshold = config['class_weights_threshold']
-    num_classes = config['num_classes']
+    class_weights_threshold = config["class_weights_threshold"]
+    num_classes = config["num_classes"]
 
     if not isinstance(class_weights_threshold, list):
         class_weights_threshold = [class_weights_threshold]
@@ -209,43 +213,72 @@ def populate_class_weights_threshold_table(location: str, config: dict, test: bo
         class_weights_threshold = [class_weights_threshold[0]] * num_classes
 
     if len(class_weights_threshold) != num_classes:
-        raise SonusAIError(f'invalid class_weights_threshold length: {len(class_weights_threshold)}')
+        raise ValueError(f"invalid class_weights_threshold length: {len(class_weights_threshold)}")
 
     con = db_connection(location=location, readonly=False, test=test)
-    con.executemany("INSERT INTO class_weights_threshold (threshold) VALUES (?)",
-                    [(item,) for item in class_weights_threshold])
+    con.executemany(
+        "INSERT INTO class_weights_threshold (threshold) VALUES (?)",
+        [(item,) for item in class_weights_threshold],
+    )
     con.commit()
     con.close()
 
 
 def populate_spectral_mask_table(location: str, config: dict, test: bool = False) -> None:
-    """Populate spectral_mask table
-    """
+    """Populate spectral_mask table"""
     from .config import get_spectral_masks
     from .mixdb import db_connection
 
     con = db_connection(location=location, readonly=False, test=test)
-    con.executemany("""
+    con.executemany(
+        """
     INSERT INTO spectral_mask (f_max_width, f_num, t_max_width, t_num, t_max_percent) VALUES (?, ?, ?, ?, ?)
-    """, [(item.f_max_width,
-           item.f_num,
-           item.t_max_width,
-           item.t_num,
-           item.t_max_percent) for item in get_spectral_masks(config)]
-                    )
+    """,
+        [
+            (
+                item.f_max_width,
+                item.f_num,
+                item.t_max_width,
+                item.t_num,
+                item.t_max_percent,
+            )
+            for item in get_spectral_masks(config)
+        ],
+    )
+    con.commit()
+    con.close()
+
+
+def populate_truth_parameters_table(location: str, config: dict, test: bool = False) -> None:
+    """Populate truth_parameters table"""
+    from .config import get_truth_parameters
+    from .mixdb import db_connection
+
+    con = db_connection(location=location, readonly=False, test=test)
+    con.executemany(
+        """
+    INSERT INTO truth_parameters (name, parameters) VALUES (?, ?)
+    """,
+        [
+            (
+                item.name,
+                item.parameters,
+            )
+            for item in get_truth_parameters(config)
+        ],
+    )
     con.commit()
     con.close()
 
 
 def populate_target_file_table(location: str, target_files: TargetFiles, test: bool = False) -> None:
-    """Populate target file table
-    """
+    """Populate target file table"""
     import json
     from pathlib import Path
 
     from .mixdb import db_connection
 
-    _populate_truth_setting_table(location, target_files, test)
+    _populate_truth_config_table(location, target_files, test)
     _populate_speaker_table(location, target_files, test)
 
     con = db_connection(location=location, readonly=False, test=test)
@@ -259,76 +292,106 @@ def populate_target_file_table(location: str, target_files: TargetFiles, test: b
             textgrid_metadata_tiers.add(tier)
 
         # Get truth settings for target file
-        truth_setting_ids: list[int] = []
-        for truth_setting in target_file.truth_settings:
-            cur.execute("SELECT truth_setting.id FROM truth_setting WHERE ? = truth_setting.setting",
-                        (truth_setting.to_json(),))
-            truth_setting_ids.append(cur.fetchone()[0])
+        truth_config_ids: list[int] = []
+        for name, config in target_file.truth_configs.items():
+            ts = json.dumps({"name": name} | config.to_dict())
+            cur.execute(
+                "SELECT truth_config.id FROM truth_config WHERE ? = truth_config.config",
+                (ts,),
+            )
+            truth_config_ids.append(cur.fetchone()[0])
 
         # Get speaker_id for target file
-        cur.execute("SELECT speaker.id FROM speaker WHERE ? = speaker.parent",
-                    (Path(target_file.name).parent.as_posix(),))
+        cur.execute(
+            "SELECT speaker.id FROM speaker WHERE ? = speaker.parent",
+            (Path(target_file.name).parent.as_posix(),),
+        )
         result = cur.fetchone()
         speaker_id = None
         if result is not None:
             speaker_id = result[0]
 
         # Add entry
-        cur.execute("INSERT INTO target_file (name, samples, level_type, speaker_id) VALUES (?, ?, ?, ?)",
-                    (target_file.name, target_file.samples, target_file.level_type, speaker_id))
+        cur.execute(
+            "INSERT INTO target_file (name, samples, class_indices, level_type, speaker_id) VALUES (?, ?, ?, ?, ?)",
+            (
+                target_file.name,
+                target_file.samples,
+                json.dumps(target_file.class_indices),
+                target_file.level_type,
+                speaker_id,
+            ),
+        )
         target_file_id = cur.lastrowid
-        for truth_setting_id in truth_setting_ids:
-            cur.execute("INSERT INTO target_file_truth_setting (target_file_id, truth_setting_id) VALUES (?, ?)",
-                        (target_file_id, truth_setting_id))
+        for truth_config_id in truth_config_ids:
+            cur.execute(
+                "INSERT INTO target_file_truth_config (target_file_id, truth_config_id) VALUES (?, ?)",
+                (target_file_id, truth_config_id),
+            )
 
     # Update textgrid_metadata_tiers in the top table
-    con.execute("UPDATE top SET textgrid_metadata_tiers=? WHERE top.id = ?",
-                (json.dumps(sorted(textgrid_metadata_tiers)), 1))
+    con.execute(
+        "UPDATE top SET textgrid_metadata_tiers=? WHERE top.id = ?",
+        (json.dumps(sorted(textgrid_metadata_tiers)), 1),
+    )
 
     con.commit()
     con.close()
 
 
 def populate_noise_file_table(location: str, noise_files: NoiseFiles, test: bool = False) -> None:
-    """Populate noise file table
-    """
+    """Populate noise file table"""
     from .mixdb import db_connection
 
     con = db_connection(location=location, readonly=False, test=test)
-    con.executemany("INSERT INTO noise_file (name, samples) VALUES (?, ?)",
-                    [(noise_file.name, noise_file.samples) for noise_file in noise_files])
+    con.executemany(
+        "INSERT INTO noise_file (name, samples) VALUES (?, ?)",
+        [(noise_file.name, noise_file.samples) for noise_file in noise_files],
+    )
     con.commit()
     con.close()
 
 
-def populate_impulse_response_file_table(location: str, impulse_response_files: ImpulseResponseFiles,
-                                         test: bool = False) -> None:
-    """Populate impulse response file table
-    """
+def populate_impulse_response_file_table(
+    location: str, impulse_response_files: ImpulseResponseFiles, test: bool = False
+) -> None:
+    """Populate impulse response file table"""
+    import json
+
     from .mixdb import db_connection
 
     con = db_connection(location=location, readonly=False, test=test)
-    con.executemany("INSERT INTO impulse_response_file (file) VALUES (?)",
-                    [(impulse_response_file,) for impulse_response_file in impulse_response_files])
+    con.executemany(
+        "INSERT INTO impulse_response_file (file, tags) VALUES (?, ?)",
+        [
+            (
+                impulse_response_file.file,
+                json.dumps(impulse_response_file.tags),
+            )
+            for impulse_response_file in impulse_response_files
+        ],
+    )
     con.commit()
     con.close()
 
 
 def update_mixid_width(location: str, num_mixtures: int, test: bool = False) -> None:
-    """Update the mixid width
-    """
-    from .mixdb import db_connection
+    """Update the mixid width"""
     from sonusai.utils import max_text_width
 
+    from .mixdb import db_connection
+
     con = db_connection(location=location, readonly=False, test=test)
-    con.execute("UPDATE top SET mixid_width=? WHERE top.id = ?", (max_text_width(num_mixtures), 1))
+    con.execute(
+        "UPDATE top SET mixid_width=? WHERE top.id = ?",
+        (max_text_width(num_mixtures), 1),
+    )
     con.commit()
     con.close()
 
 
 def populate_mixture_table(location: str, mixtures: Mixtures, test: bool = False) -> None:
-    """Populate mixture table
-    """
+    """Populate mixture table"""
     from .helpers import from_mixture
     from .helpers import from_target
     from .mixdb import db_connection
@@ -348,29 +411,35 @@ def populate_mixture_table(location: str, mixtures: Mixtures, test: bool = False
     # Populate mixture table
     cur = con.cursor()
     for mixture in mixtures:
-        cur.execute("""
+        cur.execute(
+            """
         INSERT INTO mixture (name, noise_file_id, noise_augmentation, noise_offset, noise_snr_gain, random_snr,
         snr, samples, spectral_mask_id, spectral_mask_seed, target_snr_gain)
         VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
-        """, from_mixture(mixture))
+        """,
+            from_mixture(mixture),
+        )
 
         mixture_id = cur.lastrowid
         for target in mixture.targets:
-            target_id = con.execute("""
+            target_id = con.execute(
+                """
             SELECT target.id
             FROM target
             WHERE ? = target.file_id AND ? = target.augmentation AND ? = target.gain
-            """, from_target(target)).fetchone()[0]
-            con.execute("INSERT INTO mixture_target (mixture_id, target_id) VALUES (?, ?)",
-                        (mixture_id, target_id))
+            """,
+                from_target(target),
+            ).fetchone()[0]
+            con.execute(
+                "INSERT INTO mixture_target (mixture_id, target_id) VALUES (?, ?)",
+                (mixture_id, target_id),
+            )
 
     con.commit()
     con.close()
 
 
-def update_mixture(mixdb: MixtureDatabase,
-                   mixture: Mixture,
-                   with_data: bool = False) -> tuple[Mixture, GenMixData]:
+def update_mixture(mixdb: MixtureDatabase, mixture: Mixture, with_data: bool = False) -> tuple[Mixture, GenMixData]:
     """Update mixture record with name and gains
 
     :param mixdb: Mixture database
@@ -391,12 +460,11 @@ def update_mixture(mixdb: MixtureDatabase,
     # Apply IR and sum targets audio before initializing the mixture SNR gains
     target_audio = get_target(mixdb, mixture, targets_audio)
 
-    mixture = _initialize_mixture_gains(mixdb=mixdb,
-                                        mixture=mixture,
-                                        target_audio=target_audio,
-                                        noise_audio=noise_audio)
+    mixture = _initialize_mixture_gains(
+        mixdb=mixdb, mixture=mixture, target_audio=target_audio, noise_audio=noise_audio
+    )
 
-    mixture.name = f'{int(mixture.name):0{mixdb.mixid_width}}.h5'
+    mixture.name = f"{int(mixture.name):0{mixdb.mixid_width}}"
 
     if not with_data:
         return mixture, GenMixData()
@@ -409,10 +477,12 @@ def update_mixture(mixdb: MixtureDatabase,
     target_audio = get_target(mixdb, mixture, targets_audio)
     mixture_audio = target_audio + noise_audio
 
-    return mixture, GenMixData(mixture=mixture_audio,
-                               targets=targets_audio,
-                               target=target_audio,
-                               noise=noise_audio)
+    return mixture, GenMixData(
+        mixture=mixture_audio,
+        targets=targets_audio,
+        target=target_audio,
+        noise=noise_audio,
+    )
 
 
 def _augmented_noise_audio(mixdb: MixtureDatabase, mixture: Mixture) -> AudioT:
@@ -439,9 +509,13 @@ def _initialize_targets_audio(mixdb: MixtureDatabase, mixture: Mixture) -> tuple
     targets_audio = []
     for target in mixture.targets:
         target_audio = mixdb.read_target_audio(target.file_id)
-        targets_audio.append(apply_augmentation(audio=target_audio,
-                                                augmentation=target.augmentation,
-                                                frame_length=mixdb.feature_step_samples))
+        targets_audio.append(
+            apply_augmentation(
+                audio=target_audio,
+                augmentation=target.augmentation,
+                frame_length=mixdb.feature_step_samples,
+            )
+        )
 
         # target_gain is used to back out the gain augmentation in order to return the target audio
         # to its normalized level when calculating truth (if needed).
@@ -458,13 +532,11 @@ def _initialize_targets_audio(mixdb: MixtureDatabase, mixture: Mixture) -> tuple
     return mixture, targets_audio
 
 
-def _initialize_mixture_gains(mixdb: MixtureDatabase,
-                              mixture: Mixture,
-                              target_audio: AudioT,
-                              noise_audio: AudioT) -> Mixture:
+def _initialize_mixture_gains(
+    mixdb: MixtureDatabase, mixture: Mixture, target_audio: AudioT, noise_audio: AudioT
+) -> Mixture:
     import numpy as np
 
-    from sonusai import SonusAIError
     from sonusai.utils import asl_p56
     from sonusai.utils import db_to_linear
 
@@ -480,19 +552,20 @@ def _initialize_mixture_gains(mixdb: MixtureDatabase,
         mixture.target_snr_gain = 1
         mixture.noise_snr_gain = 0
     else:
-        target_level_types = [target_file.level_type for target_file in
-                              [mixdb.target_file(target.file_id) for target in mixture.targets]]
+        target_level_types = [
+            target_file.level_type for target_file in [mixdb.target_file(target.file_id) for target in mixture.targets]
+        ]
         if not all(level_type == target_level_types[0] for level_type in target_level_types):
-            raise SonusAIError(f'Not all target_level_types in mixup are the same')
+            raise ValueError("Not all target_level_types in mixup are the same")
 
         level_type = target_level_types[0]
         match level_type:
-            case 'default':
+            case "default":
                 target_energy = np.mean(np.square(target_audio))
-            case 'speech':
+            case "speech":
                 target_energy = asl_p56(target_audio)
             case _:
-                raise SonusAIError(f'Unknown level_type: {level_type}')
+                raise ValueError(f"Unknown level_type: {level_type}")
 
         noise_energy = np.mean(np.square(noise_audio))
         if noise_energy == 0:
@@ -525,19 +598,20 @@ def _initialize_mixture_gains(mixdb: MixtureDatabase,
     return mixture
 
 
-def generate_mixtures(noise_mix_mode: str,
-                      augmented_targets: AugmentedTargets,
-                      target_files: TargetFiles,
-                      target_augmentations: AugmentationRules,
-                      noise_files: NoiseFiles,
-                      noise_augmentations: AugmentationRules,
-                      spectral_masks: SpectralMasks,
-                      all_snrs: list[UniversalSNRGenerator],
-                      mixups: list[int],
-                      num_classes: int,
-                      truth_mutex: bool,
-                      feature_step_samples: int,
-                      num_ir: int) -> tuple[int, int, Mixtures]:
+def generate_mixtures(
+    noise_mix_mode: str,
+    augmented_targets: AugmentedTargets,
+    target_files: TargetFiles,
+    target_augmentations: AugmentationRules,
+    noise_files: NoiseFiles,
+    noise_augmentations: AugmentationRules,
+    spectral_masks: SpectralMasks,
+    all_snrs: list[UniversalSNRGenerator],
+    mixups: list[int],
+    num_classes: int,
+    feature_step_samples: int,
+    num_ir: int,
+) -> tuple[int, int, Mixtures]:
     """Generate mixtures
 
     :param noise_mix_mode: Noise mix mode
@@ -550,72 +624,72 @@ def generate_mixtures(noise_mix_mode: str,
     :param all_snrs: List of all SNRs
     :param mixups: List of mixup values
     :param num_classes: Number of classes
-    :param truth_mutex: Truth mutex mode
     :param feature_step_samples: Number of samples in a feature step
     :param num_ir: Number of impulse response files
     :return: (Number of noise files used, number of noise samples used, list of mixture records)
     """
-    from sonusai import SonusAIError
-
-    if noise_mix_mode == 'exhaustive':
-        return _exhaustive_noise_mix(augmented_targets=augmented_targets,
-                                     target_files=target_files,
-                                     target_augmentations=target_augmentations,
-                                     noise_files=noise_files,
-                                     noise_augmentations=noise_augmentations,
-                                     spectral_masks=spectral_masks,
-                                     all_snrs=all_snrs,
-                                     mixups=mixups,
-                                     num_classes=num_classes,
-                                     truth_mutex=truth_mutex,
-                                     feature_step_samples=feature_step_samples,
-                                     num_ir=num_ir)
-
-    if noise_mix_mode == 'non-exhaustive':
-        return _non_exhaustive_noise_mix(augmented_targets=augmented_targets,
-                                         target_files=target_files,
-                                         target_augmentations=target_augmentations,
-                                         noise_files=noise_files,
-                                         noise_augmentations=noise_augmentations,
-                                         spectral_masks=spectral_masks,
-                                         all_snrs=all_snrs,
-                                         mixups=mixups,
-                                         num_classes=num_classes,
-                                         truth_mutex=truth_mutex,
-                                         feature_step_samples=feature_step_samples,
-                                         num_ir=num_ir)
-
-    if noise_mix_mode == 'non-combinatorial':
-        return _non_combinatorial_noise_mix(augmented_targets=augmented_targets,
-                                            target_files=target_files,
-                                            target_augmentations=target_augmentations,
-                                            noise_files=noise_files,
-                                            noise_augmentations=noise_augmentations,
-                                            spectral_masks=spectral_masks,
-                                            all_snrs=all_snrs,
-                                            mixups=mixups,
-                                            num_classes=num_classes,
-                                            truth_mutex=truth_mutex,
-                                            feature_step_samples=feature_step_samples,
-                                            num_ir=num_ir)
-
-    raise SonusAIError(f'invalid noise_mix_mode: {noise_mix_mode}')
-
-
-def _exhaustive_noise_mix(augmented_targets: AugmentedTargets,
-                          target_files: TargetFiles,
-                          target_augmentations: AugmentationRules,
-                          noise_files: NoiseFiles,
-                          noise_augmentations: AugmentationRules,
-                          spectral_masks: SpectralMasks,
-                          all_snrs: list[UniversalSNRGenerator],
-                          mixups: list[int],
-                          num_classes: int,
-                          truth_mutex: bool,
-                          feature_step_samples: int,
-                          num_ir: int) -> tuple[int, int, Mixtures]:
-    """ Use every noise/augmentation with every target/augmentation
-    """
+    if noise_mix_mode == "exhaustive":
+        return _exhaustive_noise_mix(
+            augmented_targets=augmented_targets,
+            target_files=target_files,
+            target_augmentations=target_augmentations,
+            noise_files=noise_files,
+            noise_augmentations=noise_augmentations,
+            spectral_masks=spectral_masks,
+            all_snrs=all_snrs,
+            mixups=mixups,
+            num_classes=num_classes,
+            feature_step_samples=feature_step_samples,
+            num_ir=num_ir,
+        )
+
+    if noise_mix_mode == "non-exhaustive":
+        return _non_exhaustive_noise_mix(
+            augmented_targets=augmented_targets,
+            target_files=target_files,
+            target_augmentations=target_augmentations,
+            noise_files=noise_files,
+            noise_augmentations=noise_augmentations,
+            spectral_masks=spectral_masks,
+            all_snrs=all_snrs,
+            mixups=mixups,
+            num_classes=num_classes,
+            feature_step_samples=feature_step_samples,
+            num_ir=num_ir,
+        )
+
+    if noise_mix_mode == "non-combinatorial":
+        return _non_combinatorial_noise_mix(
+            augmented_targets=augmented_targets,
+            target_files=target_files,
+            target_augmentations=target_augmentations,
+            noise_files=noise_files,
+            noise_augmentations=noise_augmentations,
+            spectral_masks=spectral_masks,
+            all_snrs=all_snrs,
+            mixups=mixups,
+            num_classes=num_classes,
+            feature_step_samples=feature_step_samples,
+            num_ir=num_ir,
+        )
+
+    raise ValueError(f"invalid noise_mix_mode: {noise_mix_mode}")
+
+
+def _exhaustive_noise_mix(
+    augmented_targets: AugmentedTargets,
+    target_files: TargetFiles,
+    target_augmentations: AugmentationRules,
+    noise_files: NoiseFiles,
+    noise_augmentations: AugmentationRules,
+    spectral_masks: SpectralMasks,
+    all_snrs: list[UniversalSNRGenerator],
+    mixups: list[int],
+    num_classes: int,
+    feature_step_samples: int,
+    num_ir: int,
+) -> tuple[int, int, Mixtures]:
+    """Use every noise/augmentation with every target/augmentation"""
     from random import randint
 
     import numpy as np
@@ -633,42 +707,53 @@ def _exhaustive_noise_mix(augmented_targets: AugmentedTargets,
     used_noise_files = len(noise_files) * len(noise_augmentations)
     used_noise_samples = 0
 
-    augmented_target_ids_for_mixups = [get_augmented_target_ids_for_mixup(augmented_targets=augmented_targets,
-                                                                          targets=target_files,
-                                                                          target_augmentations=target_augmentations,
-                                                                          mixup=mixup,
-                                                                          num_classes=num_classes,
-                                                                          truth_mutex=truth_mutex) for mixup in mixups]
+    augmented_target_ids_for_mixups = [
+        get_augmented_target_ids_for_mixup(
+            augmented_targets=augmented_targets,
+            targets=target_files,
+            target_augmentations=target_augmentations,
+            mixup=mixup,
+            num_classes=num_classes,
+        )
+        for mixup in mixups
+    ]
     for noise_file_id in range(len(noise_files)):
         for noise_augmentation_rule in noise_augmentations:
             noise_augmentation = augmentation_from_rule(noise_augmentation_rule, num_ir)
             noise_offset = 0
             noise_length = estimate_augmented_length_from_length(
                 length=noise_files[noise_file_id].samples,
-                tempo=noise_augmentation.tempo)
+                tempo=noise_augmentation.tempo,
+            )
 
             for augmented_target_ids_for_mixup in augmented_target_ids_for_mixups:
                 for augmented_target_ids in augmented_target_ids_for_mixup:
-                    targets, target_length = _get_target_info(augmented_target_ids=augmented_target_ids,
-                                                              augmented_targets=augmented_targets,
-                                                              target_files=target_files,
-                                                              target_augmentations=target_augmentations,
-                                                              feature_step_samples=feature_step_samples,
-                                                              num_ir=num_ir)
+                    targets, target_length = _get_target_info(
+                        augmented_target_ids=augmented_target_ids,
+                        augmented_targets=augmented_targets,
+                        target_files=target_files,
+                        target_augmentations=target_augmentations,
+                        feature_step_samples=feature_step_samples,
+                        num_ir=num_ir,
+                    )
 
                     for spectral_mask_id in range(len(spectral_masks)):
                         for snr in all_snrs:
-                            mixtures.append(Mixture(
-                                targets=targets,
-                                name=str(m_id),
-                                noise=Noise(file_id=noise_file_id + 1,
-                                            augmentation=noise_augmentation,
-                                            offset=noise_offset),
-                                samples=target_length,
-                                snr=UniversalSNR(value=snr.value,
-                                                 is_random=snr.is_random),
-                                spectral_mask_id=spectral_mask_id + 1,
-                                spectral_mask_seed=randint(0, np.iinfo('i').max)))
+                            mixtures.append(
+                                Mixture(
+                                    targets=targets,
+                                    name=str(m_id),
+                                    noise=Noise(
+                                        file_id=noise_file_id + 1,
+                                        augmentation=noise_augmentation,
+                                        offset=noise_offset,
+                                    ),
+                                    samples=target_length,
+                                    snr=UniversalSNR(value=snr.value, is_random=snr.is_random),
+                                    spectral_mask_id=spectral_mask_id + 1,
+                                    spectral_mask_seed=randint(0, np.iinfo("i").max),  # noqa: S311
+                                )
+                            )
                             m_id += 1
 
                             noise_offset = int((noise_offset + target_length) % noise_length)
@@ -677,19 +762,20 @@ def _exhaustive_noise_mix(augmented_targets: AugmentedTargets,
     return used_noise_files, used_noise_samples, mixtures
 
 
-def _non_exhaustive_noise_mix(augmented_targets: AugmentedTargets,
-                              target_files: TargetFiles,
-                              target_augmentations: AugmentationRules,
-                              noise_files: NoiseFiles,
-                              noise_augmentations: AugmentationRules,
-                              spectral_masks: SpectralMasks,
-                              all_snrs: list[UniversalSNRGenerator],
-                              mixups: list[int],
-                              num_classes: int,
-                              truth_mutex: bool,
-                              feature_step_samples: int,
-                              num_ir: int) -> tuple[int, int, Mixtures]:
-    """ Cycle through every target/augmentation without necessarily using all noise/augmentation combinations
+def _non_exhaustive_noise_mix(
+    augmented_targets: AugmentedTargets,
+    target_files: TargetFiles,
+    target_augmentations: AugmentationRules,
+    noise_files: NoiseFiles,
+    noise_augmentations: AugmentationRules,
+    spectral_masks: SpectralMasks,
+    all_snrs: list[UniversalSNRGenerator],
+    mixups: list[int],
+    num_classes: int,
+    feature_step_samples: int,
+    num_ir: int,
+) -> tuple[int, int, Mixtures]:
+    """Cycle through every target/augmentation without necessarily using all noise/augmentation combinations
     (reduced data set).
     """
     from random import randint
@@ -710,67 +796,81 @@ def _non_exhaustive_noise_mix(augmented_targets: AugmentedTargets,
     noise_augmentation_id = None
     noise_offset = None
 
-    augmented_target_indices_for_mixups = [get_augmented_target_ids_for_mixup(
-        augmented_targets=augmented_targets,
-        targets=target_files,
-        target_augmentations=target_augmentations,
-        mixup=mixup,
-        num_classes=num_classes,
-        truth_mutex=truth_mutex) for mixup in mixups]
+    augmented_target_indices_for_mixups = [
+        get_augmented_target_ids_for_mixup(
+            augmented_targets=augmented_targets,
+            targets=target_files,
+            target_augmentations=target_augmentations,
+            mixup=mixup,
+            num_classes=num_classes,
+        )
+        for mixup in mixups
+    ]
     for mixup in augmented_target_indices_for_mixups:
         for augmented_target_indices in mixup:
-            targets, target_length = _get_target_info(augmented_target_ids=augmented_target_indices,
-                                                      augmented_targets=augmented_targets,
-                                                      target_files=target_files,
-                                                      target_augmentations=target_augmentations,
-                                                      feature_step_samples=feature_step_samples,
-                                                      num_ir=num_ir)
+            targets, target_length = _get_target_info(
+                augmented_target_ids=augmented_target_indices,
+                augmented_targets=augmented_targets,
+                target_files=target_files,
+                target_augmentations=target_augmentations,
+                feature_step_samples=feature_step_samples,
+                num_ir=num_ir,
+            )
 
             for spectral_mask_id in range(len(spectral_masks)):
                 for snr in all_snrs:
-                    (noise_file_id,
-                     noise_augmentation_id,
-                     noise_augmentation,
-                     noise_offset) = _get_next_noise_offset(noise_file_id=noise_file_id,
-                                                            noise_augmentation_id=noise_augmentation_id,
-                                                            noise_offset=noise_offset,
-                                                            target_length=target_length,
-                                                            noise_files=noise_files,
-                                                            noise_augmentations=noise_augmentations,
-                                                            num_ir=num_ir)
+                    (
+                        noise_file_id,
+                        noise_augmentation_id,
+                        noise_augmentation,
+                        noise_offset,
+                    ) = _get_next_noise_offset(
+                        noise_file_id=noise_file_id,
+                        noise_augmentation_id=noise_augmentation_id,
+                        noise_offset=noise_offset,
+                        target_length=target_length,
+                        noise_files=noise_files,
+                        noise_augmentations=noise_augmentations,
+                        num_ir=num_ir,
+                    )
                     used_noise_samples += target_length
 
-                    used_noise_files.add(f'{noise_file_id}_{noise_augmentation_id}')
-
-                    mixtures.append(Mixture(
-                        targets=targets,
-                        name=str(m_id),
-                        noise=Noise(file_id=noise_file_id + 1,
-                                    augmentation=noise_augmentation,
-                                    offset=noise_offset),
-                        samples=target_length,
-                        snr=UniversalSNR(value=snr.value,
-                                         is_random=snr.is_random),
-                        spectral_mask_id=spectral_mask_id + 1,
-                        spectral_mask_seed=randint(0, np.iinfo('i').max)))
+                    used_noise_files.add(f"{noise_file_id}_{noise_augmentation_id}")
+
+                    mixtures.append(
+                        Mixture(
+                            targets=targets,
+                            name=str(m_id),
+                            noise=Noise(
+                                file_id=noise_file_id + 1,
+                                augmentation=noise_augmentation,
+                                offset=noise_offset,
+                            ),
+                            samples=target_length,
+                            snr=UniversalSNR(value=snr.value, is_random=snr.is_random),
+                            spectral_mask_id=spectral_mask_id + 1,
+                            spectral_mask_seed=randint(0, np.iinfo("i").max),  # noqa: S311
+                        )
+                    )
                     m_id += 1
 
     return len(used_noise_files), used_noise_samples, mixtures
 
 
-def _non_combinatorial_noise_mix(augmented_targets: AugmentedTargets,
-                                 target_files: TargetFiles,
-                                 target_augmentations: AugmentationRules,
-                                 noise_files: NoiseFiles,
-                                 noise_augmentations: AugmentationRules,
-                                 spectral_masks: SpectralMasks,
-                                 all_snrs: list[UniversalSNRGenerator],
-                                 mixups: list[int],
-                                 num_classes: int,
-                                 truth_mutex: bool,
-                                 feature_step_samples: int,
-                                 num_ir: int) -> tuple[int, int, Mixtures]:
-    """ Combine a target/augmentation with a single cut of a noise/augmentation non-exhaustively
+def _non_combinatorial_noise_mix(
+    augmented_targets: AugmentedTargets,
+    target_files: TargetFiles,
+    target_augmentations: AugmentationRules,
+    noise_files: NoiseFiles,
+    noise_augmentations: AugmentationRules,
+    spectral_masks: SpectralMasks,
+    all_snrs: list[UniversalSNRGenerator],
+    mixups: list[int],
+    num_classes: int,
+    feature_step_samples: int,
+    num_ir: int,
+) -> tuple[int, int, Mixtures]:
+    """Combine a target/augmentation with a single cut of a noise/augmentation non-exhaustively
     (each target/augmentation does not use each noise/augmentation). Cut has random start and loop back to
     beginning if end of noise/augmentation is reached.
     """
@@ -792,57 +892,72 @@ def _non_combinatorial_noise_mix(augmented_targets: AugmentedTargets,
     noise_file_id = None
     noise_augmentation_id = None
 
-    augmented_target_indices_for_mixups = [get_augmented_target_ids_for_mixup(
-        augmented_targets=augmented_targets,
-        targets=target_files,
-        target_augmentations=target_augmentations,
-        mixup=mixup,
-        num_classes=num_classes,
-        truth_mutex=truth_mutex) for mixup in mixups]
+    augmented_target_indices_for_mixups = [
+        get_augmented_target_ids_for_mixup(
+            augmented_targets=augmented_targets,
+            targets=target_files,
+            target_augmentations=target_augmentations,
+            mixup=mixup,
+            num_classes=num_classes,
+        )
+        for mixup in mixups
+    ]
     for mixup in augmented_target_indices_for_mixups:
         for augmented_target_indices in mixup:
-            targets, target_length = _get_target_info(augmented_target_ids=augmented_target_indices,
-                                                      augmented_targets=augmented_targets,
-                                                      target_files=target_files,
-                                                      target_augmentations=target_augmentations,
-                                                      feature_step_samples=feature_step_samples,
-                                                      num_ir=num_ir)
+            targets, target_length = _get_target_info(
+                augmented_target_ids=augmented_target_indices,
+                augmented_targets=augmented_targets,
+                target_files=target_files,
+                target_augmentations=target_augmentations,
+                feature_step_samples=feature_step_samples,
+                num_ir=num_ir,
+            )
 
             for spectral_mask_id in range(len(spectral_masks)):
                 for snr in all_snrs:
-                    (noise_file_id,
-                     noise_augmentation_id,
-                     noise_augmentation,
-                     noise_length) = _get_next_noise_indices(noise_file_id=noise_file_id,
-                                                             noise_augmentation_id=noise_augmentation_id,
-                                                             noise_files=noise_files,
-                                                             noise_augmentations=noise_augmentations,
-                                                             num_ir=num_ir)
+                    (
+                        noise_file_id,
+                        noise_augmentation_id,
+                        noise_augmentation,
+                        noise_length,
+                    ) = _get_next_noise_indices(
+                        noise_file_id=noise_file_id,
+                        noise_augmentation_id=noise_augmentation_id,
+                        noise_files=noise_files,
+                        noise_augmentations=noise_augmentations,
+                        num_ir=num_ir,
+                    )
                     used_noise_samples += target_length
 
-                    used_noise_files.add(f'{noise_file_id}_{noise_augmentation_id}')
-
-                    mixtures.append(Mixture(
-                        targets=targets,
-                        name=str(m_id),
-                        noise=Noise(file_id=noise_file_id + 1,
-                                    augmentation=noise_augmentation,
-                                    offset=choice(range(noise_length))),
-                        samples=target_length,
-                        snr=UniversalSNR(value=snr.value,
-                                         is_random=snr.is_random),
-                        spectral_mask_id=spectral_mask_id + 1,
-                        spectral_mask_seed=randint(0, np.iinfo('i').max)))
+                    used_noise_files.add(f"{noise_file_id}_{noise_augmentation_id}")
+
+                    mixtures.append(
+                        Mixture(
+                            targets=targets,
+                            name=str(m_id),
+                            noise=Noise(
+                                file_id=noise_file_id + 1,
+                                augmentation=noise_augmentation,
+                                offset=choice(range(noise_length)),  # noqa: S311
+                            ),
+                            samples=target_length,
+                            snr=UniversalSNR(value=snr.value, is_random=snr.is_random),
+                            spectral_mask_id=spectral_mask_id + 1,
+                            spectral_mask_seed=randint(0, np.iinfo("i").max),  # noqa: S311
+                        )
+                    )
                     m_id += 1
 
     return len(used_noise_files), used_noise_samples, mixtures
 
 
-def _get_next_noise_indices(noise_file_id: int,
-                            noise_augmentation_id: int,
-                            noise_files: NoiseFiles,
-                            noise_augmentations: AugmentationRules,
-                            num_ir: int) -> tuple[int, int, Augmentation, int]:
+def _get_next_noise_indices(
+    noise_file_id: int | None,
+    noise_augmentation_id: int | None,
+    noise_files: NoiseFiles,
+    noise_augmentations: AugmentationRules,
+    num_ir: int,
+) -> tuple[int, int, Augmentation, int]:
     from .augmentation import augmentation_from_rule
     from .augmentation import estimate_augmented_length_from_length
 
@@ -858,19 +973,21 @@ def _get_next_noise_indices(noise_file_id: int,
                 noise_file_id = 0
 
     noise_augmentation = augmentation_from_rule(noise_augmentations[noise_augmentation_id], num_ir)
-    noise_length = estimate_augmented_length_from_length(length=noise_files[noise_file_id].samples,
-                                                         tempo=noise_augmentation.tempo)
+    noise_length = estimate_augmented_length_from_length(
+        length=noise_files[noise_file_id].samples, tempo=noise_augmentation.tempo
+    )
     return noise_file_id, noise_augmentation_id, noise_augmentation, noise_length
 
 
-def _get_next_noise_offset(noise_file_id: int | None,
-                           noise_augmentation_id: int | None,
-                           noise_offset: int | None,
-                           target_length: int,
-                           noise_files: NoiseFiles,
-                           noise_augmentations: AugmentationRules,
-                           num_ir: int) -> tuple[int, int, Augmentation, int]:
-    from sonusai import SonusAIError
+def _get_next_noise_offset(
+    noise_file_id: int | None,
+    noise_augmentation_id: int | None,
+    noise_offset: int | None,
+    target_length: int,
+    noise_files: NoiseFiles,
+    noise_augmentations: AugmentationRules,
+    num_ir: int,
+) -> tuple[int, int, Augmentation, int]:
     from .augmentation import augmentation_from_rule
     from .augmentation import estimate_augmented_length_from_length
 
@@ -880,11 +997,12 @@ def _get_next_noise_offset(noise_file_id: int | None,
         noise_offset = 0
 
     noise_augmentation = augmentation_from_rule(noise_augmentations[noise_file_id], num_ir)
-    noise_length = estimate_augmented_length_from_length(length=noise_files[noise_file_id].samples,
-                                                         tempo=noise_augmentation.tempo)
+    noise_length = estimate_augmented_length_from_length(
+        length=noise_files[noise_file_id].samples, tempo=noise_augmentation.tempo
+    )
     if noise_offset + target_length >= noise_length:
         if noise_offset == 0:
-            raise SonusAIError('Length of target audio exceeds length of noise audio')
+            raise ValueError("Length of target audio exceeds length of noise audio")
 
         noise_offset = 0
         noise_augmentation_id += 1
@@ -898,12 +1016,14 @@ def _get_next_noise_offset(noise_file_id: int | None,
     return noise_file_id, noise_augmentation_id, noise_augmentation, noise_offset
 
 
-def _get_target_info(augmented_target_ids: list[int],
-                     augmented_targets: AugmentedTargets,
-                     target_files: TargetFiles,
-                     target_augmentations: AugmentationRules,
-                     feature_step_samples: int,
-                     num_ir: int) -> tuple[Targets, int]:
+def _get_target_info(
+    augmented_target_ids: list[int],
+    augmented_targets: AugmentedTargets,
+    target_files: TargetFiles,
+    target_augmentations: AugmentationRules,
+    feature_step_samples: int,
+    num_ir: int,
+) -> tuple[Targets, int]:
     from .augmentation import augmentation_from_rule
     from .augmentation import estimate_augmented_length_from_length
     from .datatypes import Target
@@ -918,18 +1038,23 @@ def _get_target_info(augmented_target_ids: list[int],
 
         mixups.append(Target(file_id=tfi + 1, augmentation=target_augmentation))
 
-        target_length = max(estimate_augmented_length_from_length(length=target_files[tfi].samples,
-                                                                  tempo=target_augmentation.tempo,
-                                                                  frame_length=feature_step_samples),
-                            target_length)
+        target_length = max(
+            estimate_augmented_length_from_length(
+                length=target_files[tfi].samples,
+                tempo=target_augmentation.tempo,
+                frame_length=feature_step_samples,
+            ),
+            target_length,
+        )
     return mixups, target_length
 
 
 def get_all_snrs_from_config(config: dict) -> list[UniversalSNRGenerator]:
     from .datatypes import UniversalSNRGenerator
 
-    return ([UniversalSNRGenerator(is_random=False, _raw_value=snr) for snr in config['snrs']] +
-            [UniversalSNRGenerator(is_random=True, _raw_value=snr) for snr in config['random_snrs']])
+    return [UniversalSNRGenerator(is_random=False, _raw_value=snr) for snr in config["snrs"]] + [
+        UniversalSNRGenerator(is_random=True, _raw_value=snr) for snr in config["random_snrs"]
+    ]
 
 
 def _get_textgrid_tiers_from_target_file(target_file: str) -> list[str]:
@@ -939,7 +1064,7 @@ def _get_textgrid_tiers_from_target_file(target_file: str) -> list[str]:
 
     from sonusai.mixture import tokenized_expand
 
-    textgrid_file = Path(tokenized_expand(target_file)[0]).with_suffix('.TextGrid')
+    textgrid_file = Path(tokenized_expand(target_file)[0]).with_suffix(".TextGrid")
     if not textgrid_file.exists():
         return []
 
@@ -949,8 +1074,7 @@ def _get_textgrid_tiers_from_target_file(target_file: str) -> list[str]:
 
 
 def _populate_speaker_table(location: str, target_files: TargetFiles, test: bool = False) -> None:
-    """Populate speaker table
-    """
+    """Populate speaker table"""
     import json
     from pathlib import Path
 
@@ -960,65 +1084,74 @@ def _populate_speaker_table(location: str, target_files: TargetFiles, test: bool
     from .tokenized_shell_vars import tokenized_expand
 
     # Determine columns for speaker table
-    all_parents = set([Path(target_file.name).parent for target_file in target_files])
-    speaker_parents = (parent for parent in all_parents if Path(tokenized_expand(parent / 'speaker.yml')[0]).exists())
+    all_parents = {Path(target_file.name).parent for target_file in target_files}
+    speaker_parents = (parent for parent in all_parents if Path(tokenized_expand(parent / "speaker.yml")[0]).exists())
 
     speakers: dict[Path, dict[str, str]] = {}
     for parent in sorted(speaker_parents):
-        with open(tokenized_expand(parent / 'speaker.yml')[0], 'r') as f:
+        with open(tokenized_expand(parent / "speaker.yml")[0]) as f:
             speakers[parent] = yaml.safe_load(f)
 
     new_columns: list[str] = []
-    for keys in speakers.keys():
-        for column in speakers[keys].keys():
+    for keys in speakers:
+        for column in speakers[keys]:
             new_columns.append(column)
     new_columns = sorted(set(new_columns))
 
     con = db_connection(location=location, readonly=False, test=test)
 
     for new_column in new_columns:
-        con.execute(f'ALTER TABLE speaker ADD COLUMN {new_column} TEXT')
+        con.execute(f"ALTER TABLE speaker ADD COLUMN {new_column} TEXT")
 
     # Populate speaker table
     speaker_rows: list[tuple[str, ...]] = []
-    for key in speakers.keys():
+    for key in speakers:
         entry = (speakers[key].get(column, None) for column in new_columns)
-        speaker_rows.append((key.as_posix(), *entry))
+        speaker_rows.append((key.as_posix(), *entry))  # type: ignore[arg-type]
 
-    column_ids = ', '.join(['parent', *new_columns])
-    column_values = ', '.join(['?'] * (len(new_columns) + 1))
-    con.executemany(f'INSERT INTO speaker ({column_ids}) VALUES ({column_values})', speaker_rows)
+    column_ids = ", ".join(["parent", *new_columns])
+    column_values = ", ".join(["?"] * (len(new_columns) + 1))
+    con.executemany(f"INSERT INTO speaker ({column_ids}) VALUES ({column_values})", speaker_rows)
 
     con.execute("CREATE INDEX speaker_parent_idx ON speaker (parent)")
 
     # Update speaker_metadata_tiers in the top table
-    tiers = [description[0] for description in con.execute("SELECT * FROM speaker").description if
-             description[0] not in ('id', 'parent')]
-    con.execute("UPDATE top SET speaker_metadata_tiers=? WHERE top.id = ?", (json.dumps(tiers), 1))
-
-    if 'speaker_id' in tiers:
+    tiers = [
+        description[0]
+        for description in con.execute("SELECT * FROM speaker").description
+        if description[0] not in ("id", "parent")
+    ]
+    con.execute(
+        "UPDATE top SET speaker_metadata_tiers=? WHERE top.id = ?",
+        (json.dumps(tiers), 1),
+    )
+
+    if "speaker_id" in tiers:
         con.execute("CREATE INDEX speaker_speaker_id_idx ON speaker (speaker_id)")
 
     con.commit()
     con.close()
 
 
-def _populate_truth_setting_table(location: str, target_files: TargetFiles, test: bool = False) -> None:
-    """Populate truth_setting table
-    """
+def _populate_truth_config_table(location: str, target_files: TargetFiles, test: bool = False) -> None:
+    """Populate truth_config table"""
+    import json
+
     from .mixdb import db_connection
 
     con = db_connection(location=location, readonly=False, test=test)
 
-    # Populate truth_setting table
-    truth_settings: list[str] = []
-    for truth_setting in [truth_setting for target_file in target_files
-                          for truth_setting in target_file.truth_settings]:
-        ts = truth_setting.to_json()
-        if ts not in truth_settings:
-            truth_settings.append(ts)
-    con.executemany("INSERT INTO truth_setting (setting) VALUES (?)",
-                    [(item,) for item in truth_settings])
+    # Populate truth_config table
+    truth_configs: list[str] = []
+    for target_file in target_files:
+        for name, config in target_file.truth_configs.items():
+            ts = json.dumps({"name": name} | config.to_dict())
+            if ts not in truth_configs:
+                truth_configs.append(ts)
+    con.executemany(
+        "INSERT INTO truth_config (config) VALUES (?)",
+        [(item,) for item in truth_configs],
+    )
 
     con.commit()
     con.close()