sonusai 0.19.8__py3-none-any.whl → 0.19.10__py3-none-any.whl

sonusai/mixture/mixdb.py

@@ -61,7 +61,7 @@ def db_connection(
     if not create and readonly:
         name += "?mode=ro"
 
-    connection = sqlite3.connect("file:" + name, uri=True)
+    connection = sqlite3.connect("file:" + name, uri=True, timeout=20)
 
     if verbose:
         connection.set_trace_callback(print)
@@ -84,7 +84,7 @@ class SQLiteContextManager:
 
 
 class MixtureDatabase:
-    def __init__(self, location: str, test: bool = False) -> None:
+    def __init__(self, location: str, test: bool = False, use_cache: bool = True) -> None:
         import json
         from os.path import exists
 
@@ -92,6 +92,7 @@ class MixtureDatabase:
 
         self.location = location
         self.test = test
+        self.use_cache = use_cache
 
         if not exists(db_file(self.location, self.test)):
             raise OSError(f"Could not find mixture database in {self.location}")
@@ -121,7 +122,7 @@ class MixtureDatabase:
             class_weights_threshold=self.class_weights_thresholds,
             feature=self.feature,
             impulse_response_files=self.impulse_response_files,
-            mixtures=self.mixtures,
+            mixtures=self.mixtures(),
             noise_mix_mode=self.noise_mix_mode,
             noise_files=self.noise_files,
             num_classes=self.num_classes,
@@ -488,7 +489,7 @@ class MixtureDatabase:
             return truth_configs
 
     def target_truth_configs(self, t_id: int) -> TruthConfigs:
-        return _target_truth_configs(self.db, t_id)
+        return _target_truth_configs(self.db, t_id, self.use_cache)
 
     @cached_property
     def random_snrs(self) -> list[float]:
@@ -556,7 +557,7 @@ class MixtureDatabase:
         :param sm_id: Spectral mask ID
         :return: Spectral mask
         """
-        return _spectral_mask(self.db, sm_id)
+        return _spectral_mask(self.db, sm_id, self.use_cache)
 
     @cached_property
     def target_files(self) -> list[TargetFile]:
@@ -619,7 +620,7 @@ class MixtureDatabase:
         :param t_id: Target file ID
         :return: Target file
         """
-        return _target_file(self.db, t_id)
+        return _target_file(self.db, t_id, self.use_cache)
 
     @cached_property
     def num_target_files(self) -> int:
@@ -657,7 +658,7 @@ class MixtureDatabase:
         :param n_id: Noise file ID
         :return: Noise file
         """
-        return _noise_file(self.db, n_id)
+        return _noise_file(self.db, n_id, self.use_cache)
 
     @cached_property
     def num_noise_files(self) -> int:
@@ -706,7 +707,7 @@ class MixtureDatabase:
         """
         if ir_id is None:
             return None
-        return _impulse_response_file(self.db, ir_id)
+        return _impulse_response_file(self.db, ir_id, self.use_cache)
 
     @cached_property
     def num_impulse_response_files(self) -> int:
@@ -717,7 +718,6 @@ class MixtureDatabase:
         with self.db() as c:
             return int(c.execute("SELECT count(impulse_response_file.id) FROM impulse_response_file").fetchone()[0])
 
-    @cached_property
     def mixtures(self) -> list[Mixture]:
         """Get mixtures from db
 
@@ -760,7 +760,7 @@ class MixtureDatabase:
         :param m_id: Zero-based mixture ID
         :return: Mixture record
         """
-        return _mixture(self.db, m_id)
+        return _mixture(self.db, m_id, self.use_cache)
 
     @cached_property
     def mixid_width(self) -> int:
@@ -805,7 +805,7 @@ class MixtureDatabase:
         """
         from .audio import read_audio
 
-        return read_audio(self.target_file(t_id).name)
+        return read_audio(self.target_file(t_id).name, self.use_cache)
 
     def augmented_noise_audio(self, mixture: Mixture) -> AudioT:
         """Get augmented noise audio
@@ -819,12 +819,12 @@ class MixtureDatabase:
         from .augmentation import apply_impulse_response
 
         noise = self.noise_file(mixture.noise.file_id)
-        audio = read_audio(noise.name)
+        audio = read_audio(noise.name, self.use_cache)
         audio = apply_augmentation(audio, mixture.noise.augmentation)
         if mixture.noise.augmentation.ir is not None:
             audio = apply_impulse_response(
                 audio,
-                read_ir(self.impulse_response_file(mixture.noise.augmentation.ir)),
+                read_ir(self.impulse_response_file(mixture.noise.augmentation.ir), self.use_cache),  # pyright: ignore [reportArgumentType]
             )
 
         return audio
@@ -1332,7 +1332,7 @@ class MixtureDatabase:
         return sorted(set(self.speaker_metadata_tiers + self.textgrid_metadata_tiers))
 
     def speaker(self, s_id: int | None, tier: str) -> str | None:
-        return _speaker(self.db, s_id, tier)
+        return _speaker(self.db, s_id, tier, self.use_cache)
 
     def speech_metadata(self, tier: str) -> list[str]:
         from .helpers import get_textgrid_tier_from_target_file
@@ -1464,7 +1464,7 @@ class MixtureDatabase:
 
         return sorted(result)
 
-    def mixture_metrics(self, m_id: int, metrics: list[str], force: bool = False) -> list[Any]:
+    def mixture_metrics(self, m_id: int, metrics: list[str], force: bool = False) -> dict[str, Any]:
         """Get metrics data for the given mixture ID
 
         :param m_id: Zero-based mixture ID
@@ -1916,21 +1916,34 @@ class MixtureDatabase:
 
             raise AttributeError(f"Unrecognized metric: '{m}'")
 
-        result: list[float | int | str | Segsnr | None] = []
+        result: dict[str, Any] = {}
         for metric in metrics:
-            result.append(calc(metric))
+            result[metric] = calc(metric)
+
+            # Check for metrics dependencies and add them even if not explicitly requested.
+            if metric.startswith("mxwer"):
+                dependencies = ("mxasr." + metric[6:], "tasr." + metric[6:])
+                for dependency in dependencies:
+                    result[dependency] = calc(dependency)
 
         return result
 
 
-@lru_cache
-def _spectral_mask(db: partial, sm_id: int) -> SpectralMask:
+def _spectral_mask(db: partial, sm_id: int, use_cache: bool = True) -> SpectralMask:
     """Get spectral mask with ID from db
 
     :param db: Database context
     :param sm_id: Spectral mask ID
+    :param use_cache: If true, use LRU caching
     :return: Spectral mask
     """
+    if use_cache:
+        return __spectral_mask(db, sm_id)
+    return __spectral_mask.__wrapped__(db, sm_id)
+
+
+@lru_cache
+def __spectral_mask(db: partial, sm_id: int) -> SpectralMask:
     from .db_datatypes import SpectralMaskRecord
 
     with db() as c:
@@ -1953,12 +1966,26 @@ def _spectral_mask(db: partial, sm_id: int) -> SpectralMask:
         )
 
 
+def _target_file(db: partial, t_id: int, use_cache: bool = True) -> TargetFile:
+    """Get target file with ID from db
+
+    :param db: Database context
+    :param t_id: Target file ID
+    :param use_cache: If true, use LRU caching
+    :return: Target file
+    """
+    if use_cache:
+        return __target_file(db, t_id, use_cache)
+    return __target_file.__wrapped__(db, t_id, use_cache)
+
+
 @lru_cache
-def _target_file(db: partial, t_id: int) -> TargetFile:
+def __target_file(db: partial, t_id: int, use_cache: bool = True) -> TargetFile:
     """Get target file with ID from db
 
     :param db: Database context
     :param t_id: Target file ID
+    :param use_cache: If true, use LRU caching
     :return: Target file
     """
     import json
@@ -1982,19 +2009,26 @@ def _target_file(db: partial, t_id: int) -> TargetFile:
             samples=target_file.samples,
             class_indices=json.loads(target_file.class_indices),
             level_type=target_file.level_type,
-            truth_configs=_target_truth_configs(db, t_id),
+            truth_configs=_target_truth_configs(db, t_id, use_cache),
             speaker_id=target_file.speaker_id,
         )
 
 
-@lru_cache
-def _noise_file(db: partial, n_id: int) -> NoiseFile:
+def _noise_file(db: partial, n_id: int, use_cache: bool = True) -> NoiseFile:
     """Get noise file with ID from db
 
     :param db: Database context
     :param n_id: Noise file ID
+    :param use_cache: If true, use LRU caching
     :return: Noise file
     """
+    if use_cache:
+        return __noise_file(db, n_id)
+    return __noise_file.__wrapped__(db, n_id)
+
+
+@lru_cache
+def __noise_file(db: partial, n_id: int) -> NoiseFile:
     with db() as c:
         noise = c.execute(
             """
@@ -2007,14 +2041,21 @@ def _noise_file(db: partial, n_id: int) -> NoiseFile:
         return NoiseFile(name=noise[0], samples=noise[1])
 
 
-@lru_cache
-def _impulse_response_file(db: partial, ir_id: int) -> str:
+def _impulse_response_file(db: partial, ir_id: int, use_cache: bool = True) -> str:
     """Get impulse response file with ID from db
 
     :param db: Database context
     :param ir_id: Impulse response file ID
-    :return: Noise
+    :param use_cache: If true, use LRU caching
+    :return: Impulse response
     """
+    if use_cache:
+        return __impulse_response_file(db, ir_id)
+    return __impulse_response_file.__wrapped__(db, ir_id)
+
+
+@lru_cache
+def __impulse_response_file(db: partial, ir_id: int) -> str:
     with db() as c:
         return str(
             c.execute(
@@ -2028,14 +2069,21 @@ def _impulse_response_file(db: partial, ir_id: int) -> str:
         )
 
 
-@lru_cache
-def _mixture(db: partial, m_id: int) -> Mixture:
+def _mixture(db: partial, m_id: int, use_cache: bool = True) -> Mixture:
     """Get mixture record with ID from db
 
     :param db: Database context
     :param m_id: Zero-based mixture ID
+    :param use_cache: If true, use LRU caching
     :return: Mixture record
     """
+    if use_cache:
+        return __mixture(db, m_id)
+    return __mixture.__wrapped__(db, m_id)
+
+
+@lru_cache
+def __mixture(db: partial, m_id: int) -> Mixture:
     from .db_datatypes import MixtureRecord
     from .db_datatypes import TargetRecord
     from .helpers import to_mixture
@@ -2068,8 +2116,14 @@ def _mixture(db: partial, m_id: int) -> Mixture:
         return to_mixture(mixture, targets)
 
 
+def _speaker(db: partial, s_id: int | None, tier: str, use_cache: bool = True) -> str | None:
+    if use_cache:
+        return __speaker(db, s_id, tier)
+    return __speaker.__wrapped__(db, s_id, tier)
+
+
 @lru_cache
-def _speaker(db: partial, s_id: int | None, tier: str) -> str | None:
+def __speaker(db: partial, s_id: int | None, tier: str) -> str | None:
     if s_id is None:
         return None
 
@@ -2082,8 +2136,14 @@ def _speaker(db: partial, s_id: int | None, tier: str) -> str | None:
         return data[0]
 
 
+def _target_truth_configs(db: partial, t_id: int, use_cache: bool = True) -> TruthConfigs:
+    if use_cache:
+        return __target_truth_configs(db, t_id)
+    return __target_truth_configs.__wrapped__(db, t_id)
+
+
 @lru_cache
-def _target_truth_configs(db: partial, t_id: int) -> TruthConfigs:
+def __target_truth_configs(db: partial, t_id: int) -> TruthConfigs:
     import json
 
     from .datatypes import TruthConfig

sonusai/mixture/torchaudio_audio.py

@@ -4,10 +4,16 @@ from sonusai.mixture.datatypes import AudioT
 from sonusai.mixture.datatypes import ImpulseResponseData
 
 
-def read_impulse_response(name: str | Path) -> ImpulseResponseData:
+def read_impulse_response(
+    name: str | Path,
+    delay_compensation: bool = True,
+    normalize: bool = True,
+) -> ImpulseResponseData:
     """Read impulse response data using torchaudio
 
     :param name: File name
+    :param delay_compensation: Apply delay compensation
+    :param normalize: Apply normalization
     :return: ImpulseResponseData object
     """
     import numpy as np
@@ -28,14 +34,19 @@ def read_impulse_response(name: str | Path) -> ImpulseResponseData:
             raise OSError(f"Error reading {name}: {e}") from e
 
     raw = torch.squeeze(raw[0, :])
-    offset = torch.argmax(raw)
-    raw = raw[offset:]
-    # Inexplicably, torch.linalg.vector_norm() causes multiprocessing contexts to hang.
-    # Use np.linalg.norm() instead.
-    # raw = raw / torch.linalg.vector_norm(raw)
+
+    if delay_compensation:
+        offset = torch.argmax(raw)
+        raw = raw[offset:]
 
     data = np.array(raw).astype(np.float32)
-    data = data / np.linalg.norm(data)
+
+    if normalize:
+        # Inexplicably,
+        #   data = data / torch.linalg.vector_norm(data)
+        # causes multiprocessing contexts to hang.
+        # Use np.linalg.norm() instead.
+        data = data / np.linalg.norm(data)
 
     return ImpulseResponseData(name=str(name), sample_rate=sample_rate, data=data)
 

sonusai/mixture/torchaudio_augmentation.py

@@ -20,10 +20,9 @@ def apply_augmentation(audio: AudioT, augmentation: Augmentation, frame_length:
 
     effects: list[list[str]] = []
 
-    # TODO
-    #  Always normalize and remove normalize from list of available augmentations
-    #  Normalize to globally set level (should this be a global config parameter,
-    #  or hard-coded into the script?)
+    # TODO: Always normalize and remove normalize from list of available augmentations
+    # Normalize to globally set level (should this be a global config parameter, or hard-coded into the script?)
+    # TODO: Support all sox effects supported by torchaudio (torchaudio.sox_effects.effect_names())
     if augmentation.normalize is not None:
         effects.append(["norm", str(augmentation.normalize)])
 

sonusai/mixture/truth_functions/energy.py

@@ -13,6 +13,7 @@ def _core(
     parameters: int,
     mapped: bool,
     snr: bool,
+    use_cache: bool = True,
 ) -> Truth:
     from os.path import join
 
@@ -50,8 +51,8 @@ def _core(
         tmp = np.nan_to_num(tmp, nan=-np.inf, posinf=np.inf, neginf=-np.inf)
 
         if mapped:
-            snr_db_mean = load_object(join(mixdb.location, config["snr_db_mean"]))
-            snr_db_std = load_object(join(mixdb.location, config["snr_db_std"]))
+            snr_db_mean = load_object(join(mixdb.location, config["snr_db_mean"]), use_cache)
+            snr_db_std = load_object(join(mixdb.location, config["snr_db_std"]), use_cache)
             tmp = _calculate_mapped_snr_f(tmp, snr_db_mean, snr_db_std)
 
         truth[frame] = tmp
@@ -85,7 +86,7 @@ def energy_f_parameters(feature: str, _num_classes: int, _config: dict) -> int:
     return ForwardTransform(**feature_forward_transform_config(feature)).bins
 
 
-def energy_f(mixdb: MixtureDatabase, m_id: int, target_index: int, config: dict) -> Truth:
+def energy_f(mixdb: MixtureDatabase, m_id: int, target_index: int, config: dict, use_cache: bool = True) -> Truth:
     """Frequency domain energy truth generation function
 
     Calculates the true energy per bin:
@@ -104,6 +105,7 @@ def energy_f(mixdb: MixtureDatabase, m_id: int, target_index: int, config: dict)
         parameters=energy_f_parameters(mixdb.feature, mixdb.num_classes, config),
         mapped=False,
         snr=False,
+        use_cache=use_cache,
     )
 
 
@@ -118,7 +120,7 @@ def snr_f_parameters(feature: str, _num_classes: int, _config: dict) -> int:
     return ForwardTransform(**feature_forward_transform_config(feature)).bins
 
 
-def snr_f(mixdb: MixtureDatabase, m_id: int, target_index: int, config: dict) -> Truth:
+def snr_f(mixdb: MixtureDatabase, m_id: int, target_index: int, config: dict, use_cache: bool = True) -> Truth:
     """Frequency domain SNR truth function documentation
 
     Calculates the true SNR per bin:
@@ -137,6 +139,7 @@ def snr_f(mixdb: MixtureDatabase, m_id: int, target_index: int, config: dict) ->
         parameters=snr_f_parameters(mixdb.feature, mixdb.num_classes, config),
         mapped=False,
         snr=True,
+        use_cache=use_cache,
     )
 
 
@@ -156,7 +159,7 @@ def mapped_snr_f_parameters(feature: str, _num_classes: int, _config: dict) -> i
     return ForwardTransform(**feature_forward_transform_config(feature)).bins
 
 
-def mapped_snr_f(mixdb: MixtureDatabase, m_id: int, target_index: int, config: dict) -> Truth:
+def mapped_snr_f(mixdb: MixtureDatabase, m_id: int, target_index: int, config: dict, use_cache: bool = True) -> Truth:
     """Frequency domain mapped SNR truth function documentation
 
     Output shape: [:, bins]
@@ -169,6 +172,7 @@ def mapped_snr_f(mixdb: MixtureDatabase, m_id: int, target_index: int, config: d
         parameters=mapped_snr_f_parameters(mixdb.feature, mixdb.num_classes, config),
         mapped=True,
         snr=True,
+        use_cache=use_cache,
     )
 
 

sonusai/mixture/truth_functions/metrics.py

@@ -25,4 +25,4 @@ def metrics(mixdb: MixtureDatabase, m_id: int, target_index: int, config: dict)
         m = [config["metric"]]
     else:
         m = config["metric"]
-    return mixdb.mixture_metrics(m_id, m)[0][target_index]
+    return mixdb.mixture_metrics(m_id, m)[m[0]][target_index]

sonusai/mkwav.py

@@ -63,7 +63,7 @@ def _process_mixture(m_id: int, location: str, write_target: bool, write_targets
     if write_noise:
         write_audio(name=join(location, "noise.wav"), audio=float_to_int16(mixdb.mixture_noise(m_id)))
 
-    write_mixture_metadata(mixdb, m_id)
+    write_mixture_metadata(mixdb, m_id=m_id)
 
 
 def main() -> None:

sonusai/onnx_predict.py

@@ -193,7 +193,7 @@ def main() -> None:
             # run inference, ort session wants i.e. batch x timesteps x feat_params, outputs numpy BxTxFP or BxFP
             predict = session.run(out_names, {in0name: feature})[0]
             # predict, _ = reshape_outputs(predict=predict[0], timesteps=frames)  # frames x feat_params
-            output_fname = join(output_dir, mixdb.mixtures[mixid].name)
+            output_fname = join(output_dir, mixdb.mixture(mixid).name)
             with h5py.File(output_fname, "a") as f:
                 if "predict" in f:
                     del f["predict"]

sonusai/queries/queries.py

@@ -178,7 +178,7 @@ def get_mixids_from_snr(
     result: dict[float, list[int]] = {}
     for snr in snrs:
         # Get a list of mixids for each SNR
-        result[snr] = sorted([i for i, mixture in enumerate(mixdb.mixtures) if mixture.snr == snr and i in mixid_out])
+        result[snr] = sorted([i for i, mixture in enumerate(mixdb.mixtures()) if mixture.snr == snr and i in mixid_out])
 
     return result
 

sonusai/utils/asr.py

@@ -65,7 +65,7 @@ def calc_asr(audio: AudioT | str, engine: str, **config) -> ASRResult:
     from sonusai.mixture import read_audio
 
     if not isinstance(audio, np.ndarray):
-        audio = copy(read_audio(audio))
+        audio = copy(read_audio(audio, config.get("use_cache", True)))
 
     return _asr_fn(engine)(audio, **config)
 

sonusai/utils/load_object.py

@@ -2,9 +2,15 @@ from functools import lru_cache
 from typing import Any
 
 
+def load_object(name: str, use_cache: bool = True) -> Any:
+    """Load an object from a pickle file"""
+    if use_cache:
+        return _load_object(name)
+    return _load_object.__wrapped__(name)
+
+
 @lru_cache
-def load_object(name: str) -> Any:
-    """Load an object from a pickle file (with LRU caching)"""
+def _load_object(name: str) -> Any:
     import pickle
     from os.path import exists
 

sonusai/utils/stratified_shuffle_split.py

@@ -42,7 +42,7 @@ def stratified_shuffle_split_mixid(
         raise ValueError("vsplit must be between 0 and 1")
 
     a_class_mixid: dict[int, list[int]] = {i + 1: [] for i in range(mixdb.num_classes)}
-    for mixid, mixture in enumerate(mixdb.mixtures):
+    for mixid, mixture in enumerate(mixdb.mixtures()):
         class_count = get_class_count_from_mixids(mixdb, mixid)
         if any(class_count):
             for class_index in mixdb.target_files[mixture.targets[0].file_id].class_indices:

{sonusai-0.19.8.dist-info → sonusai-0.19.10.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: sonusai
-Version: 0.19.8
+Version: 0.19.10
 Summary: Framework for building deep neural network models for sound, speech, and voice AI
 Home-page: https://aaware.com
 License: GPL-3.0-only

{sonusai-0.19.8.dist-info → sonusai-0.19.10.dist-info}/RECORD

@@ -1,9 +1,11 @@
 sonusai/__init__.py,sha256=NSb0bvmAh6Rm2MDtchpAGsg8a3BrmVnShYb-vC_emH8,2802
 sonusai/aawscd_probwrite.py,sha256=QZLMQrmPr3OjZ06buyYDwlnk9YPCpyr4KHkBjPsiqjU,3700
 sonusai/audiofe.py,sha256=iFdthh4UrOvziT8urjrjD7dACWZPQz9orM5bVAW3WSQ,11269
-sonusai/calc_metric_spenh.py,sha256=NTqYBNDR6LTSdDA8s4WCSEwajE3-fNJg1TUzhNzJC9g,46719
+sonusai/calc_metric_spenh.py,sha256=XWa2DzLSCEQ6GzsJv-YHfnN51f_oFwcRMMgMzusAvYA,49304
 sonusai/data/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 sonusai/data/genmixdb.yml,sha256=U_kLbE7gZ5rA7yNSB2NW7eK5dnYP5grJVMR321VMLt8,940
+sonusai/data/silero_vad_v5.1.jit,sha256=hcSOHw7LYE5dKiaPPM-5EtT36TWs3IavWj_FsK6nspo,2269612
+sonusai/data/silero_vad_v5.1.onnx,sha256=JiOilT9v89LB5hdAxs23FoEzR5smff7xFKSjzFvdeI8,2327524
 sonusai/data/speech_ma01_01.wav,sha256=PK0vMKg-NR6rPE3KouxHGF6PKXnJCr7AwjMqfu98LUA,76644
 sonusai/data/whitenoise.wav,sha256=I2umov0m34y56F9IsIBi1XtE76ZeZaSKDf70cJRe3pI,1920044
 sonusai/deprecated/gentcst.py,sha256=nKbHy3aHreHqA-XnLQOzOApS8RuTNUFqnx52a8I5zLQ,19921
@@ -12,10 +14,10 @@ sonusai/deprecated/tplot.py,sha256=0p238DvTaP4oU9y-dp0JdLaTV4TKrooAwbx7zdz_QAc,1
 sonusai/doc/__init__.py,sha256=KyQ26Um0RM8A3GYsb_tbFH64RwpoAw6lja2f_moUWas,33
 sonusai/doc/doc.py,sha256=VZXauwbOb-VIufWw-lu0yfrd6jMRPeFeVPaaEjZNvn4,18881
 sonusai/doc.py,sha256=zSmXpioB0YS_5-7kqfS5cr--veSaXkxRKzldId9Hyoc,878
-sonusai/genft.py,sha256=TqtmexKw7tZsugU1DbZ3fY7_YZ2hFlIU-ema6f0LRgQ,5586
-sonusai/genmetrics.py,sha256=RiEYmkRl0yVCIr9PHkKRqEG68WR77pP9WpWAYeKnMLE,5723
-sonusai/genmix.py,sha256=pugp74prksf4ZiJYMXWg2l3v48Qqi9KpP2WhPEBpZOU,6702
-sonusai/genmixdb.py,sha256=QB9iARnAutuPPxML-g-kc3C-W1Sodabr68e4GSRD21Q,18680
+sonusai/genft.py,sha256=K2wjO5J48UgyhCj2Sx789nkjt0DWtYgnRDbQyNtjCSY,5591
+sonusai/genmetrics.py,sha256=jORQCdf_SCrtcvDd47lgcPgQTplG956RTAqmf58Xe8Y,5689
+sonusai/genmix.py,sha256=mSc5FfAYrUt3zloPSnp81dks8ntvSH6jyk-nh97wnww,6707
+sonusai/genmixdb.py,sha256=SsbHRpPoJ77XzOBQRRDheucyuJzE-tucQtRoYl89ApU,17841
 sonusai/lsdb.py,sha256=0HOGDDndB3LT9cz9AaxKIpt9vslAoSP4F239gply4Xg,5149
 sonusai/main.py,sha256=HbnEia1B1-Z-mlHkLfojH8aj9GIpL1Btw3oH60T_CCQ,2590
 sonusai/metrics/__init__.py,sha256=ssV6JEK_oklRSocsp6HMcG-GtJvV8IkRQtdKhHHmwU8,878
@@ -35,8 +37,9 @@ sonusai/metrics/class_summary.py,sha256=ZA7zNgwBpmTs1TP_t4jRT0pWnDnATC_up_8qE4aH
 sonusai/metrics/confusion_matrix_summary.py,sha256=zBL_Ke7wF6oKtrKZPr0fsyF_taofdjxBlZmKodu0xUA,3143
 sonusai/metrics/one_hot.py,sha256=hmuyh-9tpRjb_oyqU3WqZ14zItpRJQfcqBDKJeb5H9I,13930
 sonusai/metrics/snr_summary.py,sha256=t8Fi_8WtboTi8flkZuOiHq9H3-nIELx4AKvnm-qvxLQ,5785
-sonusai/mixture/__init__.py,sha256=ccz_UpB7Rh6wUB7uwLgNhciMZhKE9Qt9ICLyjq08kxs,5126
-sonusai/mixture/audio.py,sha256=Tg0-HzSgZDttKwWwONqgK6qK4maTSmWcfiL1atZcIXQ,2390
+sonusai/metrics_summary.py,sha256=HVqjgCavxM1yzyoeDZSg_bJaXrifNQxNY7xYNKKva8g,12004
+sonusai/mixture/__init__.py,sha256=ePkmFbBltwHsx1eJDb_RDieTceZtqa1wVY1D2Pfg2rw,5162
+sonusai/mixture/audio.py,sha256=5iq39_Q0q9xuN_FNylvnn-gAZ8Io3Ir1Mqj60mVQeaQ,3432
 sonusai/mixture/augmentation.py,sha256=s8QlPHnFJOblRU59fMQ-Zqysiv4OUJ7CxLRcV81lnaA,10407
 sonusai/mixture/class_count.py,sha256=zcC3BDYMPN6wJYmO1RcOuqmrnTQIbMSznl33oN3e2sc,597
 sonusai/mixture/config.py,sha256=g5ZmOhFYqmEdRQYSgfDIZ9VM0QiTwBqk7vIyAvxnPMo,24211
@@ -46,10 +49,10 @@ sonusai/mixture/datatypes.py,sha256=xNDBWFTVQ3plJ7qHKzrXyV4pffPYuf1xMVqBsR40n4o,
 sonusai/mixture/db_datatypes.py,sha256=kvdUOMS6Pkkj9AmxCiq6zM8x7jbPPi933tVaXRxbTdQ,1534
 sonusai/mixture/eq_rule_is_valid.py,sha256=O3gCAs_0hpxENK5b7kxxpDmOpKHlXGBWuLGT_97ARSM,1210
 sonusai/mixture/feature.py,sha256=L0bPFG0RO-CrrtTStUMt_14euYsVo8_TWTP2IKSFKaA,2335
-sonusai/mixture/generation.py,sha256=sPsl-r5d9mkSglHOG5_VbmoDNqOscuWTdu7F6OJa-Qw,38751
-sonusai/mixture/helpers.py,sha256=UEN_9LttIZwvms9nOkjMr_Yf3lpb7sMPmDVUoX8D7yQ,15225
+sonusai/mixture/generation.py,sha256=yoJOcY9KPe_B1RVnENVr4ekcnXyZJMdvKMbJggpLOi4,38084
+sonusai/mixture/helpers.py,sha256=Bt9njNb_OZ3j02qgrVEMZiL0hX4kXtFK_tkPoGoeb4Y,15787
 sonusai/mixture/log_duration_and_sizes.py,sha256=qhgl87C2KbjxLdKEpjYOoqNL6rc-8-PB4R7Gx_7UG8g,1240
-sonusai/mixture/mixdb.py,sha256=W8B54U6f-8szMylXSnU0YzOR-a9M3XoCRpyqOgBIJuQ,73498
+sonusai/mixture/mixdb.py,sha256=Yg3FQqb6oI3LsFh_00CvMeH1Rrmn2pA5waaAyJDCpfY,75912
 sonusai/mixture/soundfile_audio.py,sha256=At_ZC2b9pZ_9IYp1UxyPzRoBK9-1cKPCLMm74F1AjKE,4092
 sonusai/mixture/sox_audio.py,sha256=7ouCLqXYS6tjG2L0v5lugVO7z5UwJmsr1VigbrXhs74,16725
 sonusai/mixture/sox_augmentation.py,sha256=DtfGLPaB1BIt2wvTEA__MYkGFNU85Tuup5BFsIVrh0E,4546
@@ -57,22 +60,22 @@ sonusai/mixture/spectral_mask.py,sha256=U9XJ_SAoI9b67K_3SE7bNw6U8cPGFOBttaZAxMjA
 sonusai/mixture/target_class_balancing.py,sha256=o_TZ8kVYq10lgeXHh3GUFfflfdUvRt0FekFu2eaNkDs,4251
 sonusai/mixture/targets.py,sha256=6emo2fxxp9ZhSpHuUM9xIjYMz8zeIHAw684jT3l7fAs,6442
 sonusai/mixture/tokenized_shell_vars.py,sha256=lXTzUDutuBWGV1zIsqeIxWmy-eKm0Vx1y8-iLdsL1gQ,4921
-sonusai/mixture/torchaudio_audio.py,sha256=169VXKEoOHc5nyiHJwaihkcQ_a1ZH_O-nnk9Gq4JtaQ,2887
-sonusai/mixture/torchaudio_augmentation.py,sha256=68QpJ4JcAH2fsL4qVck7VvBv615UbCBPEnNe4HMvA_8,3844
+sonusai/mixture/torchaudio_audio.py,sha256=72Hxo5TKAW7mYpRy15QFfD7AYDORBk6bVCcHENniWGw,3116
+sonusai/mixture/torchaudio_augmentation.py,sha256=uFAKxIfs50J5FR-WXodsEACm2Ao-t5dZRSJ0DwTAfBg,3930
 sonusai/mixture/truth.py,sha256=-CwwawFRGjqodR2yKvAMGL1XaYLct-tli7wZ2gbhLtQ,2121
 sonusai/mixture/truth_functions/__init__.py,sha256=0mlOFChPnXG5BC0eKOe4n9VH17jY4iOqZFLuF6Gprdk,1505
 sonusai/mixture/truth_functions/crm.py,sha256=iidcffXfqV8k9O5wt5KTWIAFaTSjmhV5ucKZPbTgpvQ,3809
-sonusai/mixture/truth_functions/energy.py,sha256=dlxl5c58bv012KgyKeO78urXSh6FK3LdqiKk0SDkmS8,6540
+sonusai/mixture/truth_functions/energy.py,sha256=BMpyFoFDRsKEv3ZxZAJPLgMgkBkA6AtGBg3MjRu1do8,6749
 sonusai/mixture/truth_functions/file.py,sha256=pyCAhx3PhJRBoZMrjoQI4Tbi5TN7sPembSVEr80Bu3g,1431
 sonusai/mixture/truth_functions/metadata.py,sha256=aEZly5bJEaZpUBZonWvcu14_Dn3M2HamwTaM5Bg7Tm8,778
-sonusai/mixture/truth_functions/metrics.py,sha256=xfH98eXBe20kwRk-tdWbBkOplF1cCA-u_naJUiTps0Y,873
+sonusai/mixture/truth_functions/metrics.py,sha256=AzQjKJ7rihk_UXOz0Atyktpzo2g9ZPMZVZPCESBIoao,876
 sonusai/mixture/truth_functions/phoneme.py,sha256=jwBYiNwwBwh2tHtOJ2NopYWhT6y19kXzSIag0XW9GSY,778
 sonusai/mixture/truth_functions/sed.py,sha256=C0n9DkfBNQblFsFCkPbooy54KuHSY7B0f1vLft2asdw,3832
 sonusai/mixture/truth_functions/target.py,sha256=nSkHFESzCEOljcYf4jQ7FmxsAWJtMCRRWFKM_DyjoLU,4926
-sonusai/mkwav.py,sha256=xoqrX76gC1zE-Xb1Zfm9ZpcZ95l0qF1ZyOPmLMkFomk,4053
-sonusai/onnx_predict.py,sha256=Y1VUN0wuvloEW46uxg7X4ywaec_Xx92djCU3BP0KAx0,8699
+sonusai/mkwav.py,sha256=ElivON2G_BT_ffKnePmPoeydl0g2DLGrbIFxfn_I1XI,4058
+sonusai/onnx_predict.py,sha256=T97ceb9stR_QtJCA-Rmv67OIeaLdyhyCf1jQ9kVOYn8,8698
 sonusai/queries/__init__.py,sha256=bhoeOFfu9GA5DOUuxRrIev7MYdXaGN8xdKJ6BXyNNtQ,277
-sonusai/queries/queries.py,sha256=N84UcRrgDPLGJZesS2KQ9eyRzPRSyukjOlc786h5Ehs,7509
+sonusai/queries/queries.py,sha256=srcEYBqLJhjqyfuJ-FwNkUwpjxYiNQeybxL3eQGm2nw,7511
 sonusai/speech/__init__.py,sha256=vqAymCBPjMUSM4OZKHTai6BYwXsOBlf_G_vOhELVf8I,133
 sonusai/speech/l2arctic.py,sha256=VQNKuTbmlbW0PJ7bOjx9sr0VjUYxJnxfTiPJIa4OOaA,3829
 sonusai/speech/librispeech.py,sha256=ugP3NVOenSsBF1cUG4Nyl7dumGHQmE4Ugk1yYjtOyj4,3070
@@ -85,7 +88,7 @@ sonusai/speech/voxceleb.py,sha256=Uu1kB1krf8hess1yuvGbYfV_VgYhklEyoz4I7KfrVpw,26
 sonusai/summarize_metric_spenh.py,sha256=2w81ZgJahYvD6wCpE3DFoUFrXexLXjO44ITRVm1HJXw,1858
 sonusai/utils/__init__.py,sha256=z72OlzZCHpYfYHKnHn7jznj6Zt7zB-FyO6hIgFk45As,2379
 sonusai/utils/asl_p56.py,sha256=cPUVwXawF7vLJgs4zUtoRGk7Wdbe5KKti_-v_8xIU10,3862
-sonusai/utils/asr.py,sha256=20eKkatNqGrbxFQoyeCRzqXaSH3lcA-9ZXGneCXjLvs,2791
+sonusai/utils/asr.py,sha256=ubiU3E61HN3r9MhPV7ci37cnLZowll8KfjUS7os3Sho,2822
 sonusai/utils/asr_functions/__init__.py,sha256=HKGRm_c48tcxlfwqH63m-MvhAoK_pCcw76lxmFmiP_U,63
 sonusai/utils/asr_functions/aaware_whisper.py,sha256=M9Y8Pgh1oIrDOPZZPSRPDig8foxfgs3f8AsoZ8W00B0,2120
 sonusai/utils/audio_devices.py,sha256=_Eiah86SZjbdp2baD2AUVF4FmhseiNuG3KJkd_LbULk,2041
@@ -104,7 +107,7 @@ sonusai/utils/get_frames_per_batch.py,sha256=xnq4tV7MT74N0H6b5ZsiAezqdXucboCLQw1
 sonusai/utils/get_label_names.py,sha256=df4jZVaQ3WnYQqNj21iUV4aYWyQEZUNmgs93qKW-_rA,820
 sonusai/utils/grouper.py,sha256=qyZ0nj84yOrC-RZsXHC-KJvcUliGktnV8S6-P3PD6_w,203
 sonusai/utils/human_readable_size.py,sha256=DOCS7SAymrtTZli8AczvyCMCh44r7ZDgVBA7jSZupmA,356
-sonusai/utils/load_object.py,sha256=-XMMwSNUPhtg6G6FhLwL1zvxwu-PLwj25QYUEielW2k,356
+sonusai/utils/load_object.py,sha256=if4Vammcd-jZTz_n7QzwNIlN4HqSL0v91I9YQzcvEEA,493
 sonusai/utils/max_text_width.py,sha256=pxiJMwb_zlkNntexgo7S6lAuF7NLLZvFdOCkxdsQJVY,315
 sonusai/utils/model_utils.py,sha256=OIJBhOjxR0wpxsd7A2r6J2AjqfdYgZzi6UEThw4S1lI,828
 sonusai/utils/numeric_conversion.py,sha256=iFPXFU8C_1mW5tmDqHq8-xP1tL8nVaSmhQRakdCqy30,328
@@ -117,11 +120,11 @@ sonusai/utils/read_predict_data.py,sha256=PUSroxmWQGtr6_EcdSHmIFQoRGou8CKKqcggWy
 sonusai/utils/reshape.py,sha256=Ozuh3UlmAS5NCeOK7NR8KgcQacHvgq10pys0VfCnOPU,5746
 sonusai/utils/seconds_to_hms.py,sha256=9Ya9O97txFtTIXZUQw1K8g7b7Xx-ptvUtMUlzsIduTo,260
 sonusai/utils/stacked_complex.py,sha256=JW6iAa1C-4Tuh4dD5c-D-O-yo-OY5Xm0AKVU0YsqsJU,2782
-sonusai/utils/stratified_shuffle_split.py,sha256=d7WLUirywSvgZWkt_5a0F8YvTnJjuXlyPB5CRmHCN0U,6704
+sonusai/utils/stratified_shuffle_split.py,sha256=fcGW8nkZIwUqq1qtxbK_ZH58sYULqZfv7iNBQnKGH-M,6706
 sonusai/utils/write_audio.py,sha256=0lKdaX57N6H-UWdioqmXCJMjwT1eBz5B-bSGqDvloAc,838
 sonusai/utils/yes_or_no.py,sha256=0h1okjXmDNbJp7rZJFR2V-HFU1GJDm3YFTUVmYExkOU,263
 sonusai/vars.py,sha256=kBBzuvC8szmdIZEEDA7XXmD765addZKdM2aFipeGO1w,933
-sonusai-0.19.8.dist-info/METADATA,sha256=KQEqF0XgSi8xV6rSYzn-bync-wXfEjzGyWXgb909gDo,2535
-sonusai-0.19.8.dist-info/WHEEL,sha256=Nq82e9rUAnEjt98J6MlVmMCZb-t9cYE2Ir1kpBmnWfs,88
-sonusai-0.19.8.dist-info/entry_points.txt,sha256=zMNjEphEPO6B3cD1GNpit7z-yA9tUU5-j3W2v-UWstU,92
-sonusai-0.19.8.dist-info/RECORD,,
+sonusai-0.19.10.dist-info/METADATA,sha256=ibwwklSb5-vmwAJMdRhW0MBWxqQYFVsYpEx5-8oaRXI,2536
+sonusai-0.19.10.dist-info/WHEEL,sha256=Nq82e9rUAnEjt98J6MlVmMCZb-t9cYE2Ir1kpBmnWfs,88
+sonusai-0.19.10.dist-info/entry_points.txt,sha256=zMNjEphEPO6B3cD1GNpit7z-yA9tUU5-j3W2v-UWstU,92
+sonusai-0.19.10.dist-info/RECORD,,