sonusai 0.19.10__py3-none-any.whl → 0.20.2__py3-none-any.whl

sonusai/mixture/generation.py

@@ -93,7 +93,8 @@ def initialize_db(location: str, test: bool = False) -> None:
     CREATE TABLE impulse_response_file (
     id INTEGER PRIMARY KEY NOT NULL,
     file TEXT NOT NULL,
-    tags TEXT NOT NULL)
+    tags TEXT NOT NULL,
+    delay INTEGER NOT NULL)
     """)
 
     con.execute("""
@@ -360,11 +361,12 @@ def populate_impulse_response_file_table(
 
     con = db_connection(location=location, readonly=False, test=test)
     con.executemany(
-        "INSERT INTO impulse_response_file (file, tags) VALUES (?, ?)",
+        "INSERT INTO impulse_response_file (file, tags, delay) VALUES (?, ?, ?)",
         [
             (
                 impulse_response_file.file,
                 json.dumps(impulse_response_file.tags),
+                impulse_response_file.delay,
             )
             for impulse_response_file in impulse_response_files
         ],
@@ -529,17 +531,13 @@ def update_mixture(mixdb: MixtureDatabase, mixture: Mixture, with_data: bool = F
 
 def _augmented_noise_audio(mixdb: MixtureDatabase, mixture: Mixture) -> AudioT:
     from .audio import read_audio
-    from .audio import read_ir
     from .augmentation import apply_augmentation
-    from .augmentation import apply_impulse_response
 
     noise = mixdb.noise_file(mixture.noise.file_id)
     noise_augmentation = mixture.noise.augmentation
 
     audio = read_audio(noise.name)
-    audio = apply_augmentation(audio, noise_augmentation)
-    if noise_augmentation.ir is not None:
-        audio = apply_impulse_response(audio, read_ir(mixdb.impulse_response_file(noise_augmentation.ir)))  # pyright: ignore [reportArgumentType]
+    audio = apply_augmentation(mixdb, audio, noise_augmentation.pre)
 
     return audio
 
@@ -553,8 +551,9 @@ def _initialize_targets_audio(mixdb: MixtureDatabase, mixture: Mixture) -> tuple
         target_audio = mixdb.read_target_audio(target.file_id)
         targets_audio.append(
             apply_augmentation(
+                mixdb=mixdb,
                 audio=target_audio,
-                augmentation=target.augmentation,
+                augmentation=target.augmentation.pre,
                 frame_length=mixdb.feature_step_samples,
             )
         )
@@ -680,7 +679,7 @@ def _exhaustive_noise_mix(
             noise_offset = 0
             noise_length = estimate_augmented_length_from_length(
                 length=noise_files[noise_file_id].samples,
-                tempo=noise_augmentation.tempo,
+                tempo=noise_augmentation.pre.tempo,
             )
 
             for augmented_target_ids_for_mixup in augmented_target_ids_for_mixups:
@@ -922,7 +921,7 @@ def _get_next_noise_indices(
 
     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
+        length=noise_files[noise_file_id].samples, tempo=noise_augmentation.pre.tempo
     )
     return noise_file_id, noise_augmentation_id, noise_augmentation, noise_length
 
@@ -946,7 +945,7 @@ def _get_next_noise_offset(
 
     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
+        length=noise_files[noise_file_id].samples, tempo=noise_augmentation.pre.tempo
     )
     if noise_offset + target_length >= noise_length:
         if noise_offset == 0:
@@ -987,7 +986,7 @@ def _get_target_info(
         target_length = max(
             estimate_augmented_length_from_length(
                 length=target_files[tfi].samples,
-                tempo=target_augmentation.tempo,
+                tempo=target_augmentation.pre.tempo,
                 frame_length=feature_step_samples,
             ),
             target_length,

sonusai/mixture/helpers.py

@@ -117,11 +117,11 @@ def mixture_all_speech_metadata(mixdb: MixtureDatabase, mixture: Mixture) -> lis
                 # Check for tempo augmentation and adjust Interval start and end data as needed
                 entries = []
                 for entry in item:
-                    if target.augmentation.tempo is not None:
+                    if target.augmentation.pre.tempo is not None:
                         entries.append(
                             Interval(
-                                entry.start / target.augmentation.tempo,
-                                entry.end / target.augmentation.tempo,
+                                entry.start / target.augmentation.pre.tempo,
+                                entry.end / target.augmentation.pre.tempo,
                                 entry.label,
                             )
                         )
@@ -153,10 +153,8 @@ def mixture_metadata(mixdb: MixtureDatabase, m_id: int | None = None, mixture: M
     speech_metadata = mixture_all_speech_metadata(mixdb, mixture)
     for mi, target in enumerate(mixture.targets):
         target_file = mixdb.target_file(target.file_id)
-        target_augmentation = target.augmentation
         metadata += f"target {mi} name: {target_file.name}\n"
         metadata += f"target {mi} augmentation: {target.augmentation.to_dict()}\n"
-        metadata += f"target {mi} ir: {mixdb.impulse_response_file(target_augmentation.ir)}\n"
         metadata += f"target {mi} target_gain: {target.gain if not mixture.is_noise_only else 0}\n"
         metadata += f"target {mi} class indices: {target_file.class_indices}\n"
         for key in target_file.truth_configs:
@@ -168,7 +166,6 @@ def mixture_metadata(mixdb: MixtureDatabase, m_id: int | None = None, mixture: M
     noise_augmentation = mixture.noise.augmentation
     metadata += f"noise name: {noise.name}\n"
     metadata += f"noise augmentation: {noise_augmentation.to_dict()}\n"
-    metadata += f"noise ir: {mixdb.impulse_response_file(noise_augmentation.ir)}\n"
     metadata += f"noise offset: {mixture.noise_offset}\n"
     metadata += f"snr: {mixture.snr}\n"
     metadata += f"random_snr: {mixture.snr.is_random}\n"
@@ -260,33 +257,33 @@ def to_target(entry: TargetRecord) -> Target:
     )
 
 
-def get_target(mixdb: MixtureDatabase, mixture: Mixture, targets_audio: list[AudioT], use_cache: bool = True) -> AudioT:
+def get_target(mixdb: MixtureDatabase, mixture: Mixture, targets_audio: list[AudioT]) -> AudioT:
     """Get the augmented target audio data for the given mixture record
 
     :param mixdb: Mixture database
     :param mixture: Mixture record
     :param targets_audio: List of augmented target audio data (one per target in the mixup)
-    :param use_cache: If true, use LRU caching
     :return: Sum of augmented target audio data
     """
-    # Apply impulse responses to targets
+    # Apply post-truth augmentation effects to targets and sum
     import numpy as np
 
-    from .audio import read_ir
-    from .augmentation import apply_impulse_response
-
-    targets_ir = []
-    for idx, target in enumerate(targets_audio):
-        ir_idx = mixture.targets[idx].augmentation.ir
-        if ir_idx is not None:
-            targets_ir.append(
-                apply_impulse_response(audio=target, ir=read_ir(mixdb.impulse_response_file(int(ir_idx)), use_cache))  # pyright: ignore [reportArgumentType]
+    from .augmentation import apply_augmentation
+
+    targets_post = []
+    for idx, target_audio in enumerate(targets_audio):
+        target = mixture.targets[idx]
+        targets_post.append(
+            apply_augmentation(
+                mixdb=mixdb,
+                audio=target_audio,
+                augmentation=target.augmentation.post,
+                frame_length=mixdb.feature_step_samples,
             )
-        else:
-            targets_ir.append(target)
+        )
 
     # Return sum of targets
-    return np.sum(targets_ir, axis=0)
+    return np.sum(targets_post, axis=0)
 
 
 def get_transform_from_audio(audio: AudioT, transform: ForwardTransform) -> tuple[AudioF, EnergyT]:
@@ -400,7 +397,7 @@ def augmented_target_samples(
         [
             estimate_augmented_length_from_length(
                 length=target_files[fi].samples,
-                tempo=target_augmentations[ai].tempo,
+                tempo=target_augmentations[ai].pre.tempo,
                 frame_length=feature_step_samples,
             )
             for fi, ai in it
@@ -420,7 +417,7 @@ def augmented_noise_samples(noise_files: list[NoiseFile], noise_augmentations: l
 def augmented_noise_length(noise_file: NoiseFile, noise_augmentation: Augmentation) -> int:
     from .augmentation import estimate_augmented_length_from_length
 
-    return estimate_augmented_length_from_length(length=noise_file.samples, tempo=noise_augmentation.tempo)
+    return estimate_augmented_length_from_length(length=noise_file.samples, tempo=noise_augmentation.pre.tempo)
 
 
 def get_textgrid_tier_from_target_file(target_file: str, tier: str) -> SpeechMetadata | None:

sonusai/mixture/ir_delay.py

@@ -0,0 +1,63 @@
+import numpy as np
+
+
+def get_impulse_response_delay(file: str) -> int:
+    from sonusai.utils import temp_seed
+
+    from .audio import raw_read_audio
+
+    ir, sample_rate = raw_read_audio(file)
+
+    with temp_seed(42):
+        wgn_ref = np.random.normal(loc=0, scale=0.2, size=int(np.ceil(0.05 * sample_rate))).astype(np.float32)
+
+    wgn_conv = np.convolve(ir, wgn_ref)
+
+    return int(np.round(tdoa(wgn_conv, wgn_ref, interp=16, phat=True)))
+
+
+def tdoa(signal: np.ndarray, reference: np.ndarray, interp: int = 1, phat: bool = False, fs: int | float = 1) -> float:
+    """Estimates the shift of array signal with respect to reference using generalized cross-correlation.
+
+    :param signal: The array whose tdoa is measured
+    :param reference: The reference array
+    :param interp: Interpolation factor for the output array
+    :param phat: Apply the PHAT weighting
+    :param fs: The sampling frequency of the input arrays
+    :return: The estimated delay between the two arrays
+    """
+    n_reference = reference.shape[0]
+
+    r_12 = correlate(signal, reference, interp=interp, phat=phat)
+
+    delay = (np.argmax(np.abs(r_12)) / interp - (n_reference - 1)) / fs
+
+    return float(delay)
+
+
+def correlate(x1: np.ndarray, x2: np.ndarray, interp: int = 1, phat: bool = False) -> np.ndarray:
+    """Compute the cross-correlation between x1 and x2
+
+    :param x1: Input array 1
+    :param x2: Input array 2
+    :param interp: Interpolation factor for the output array
+    :param phat: Apply the PHAT weighting
+    :return: The cross-correlation between the two arrays
+    """
+    n_x1 = x1.shape[0]
+    n_x2 = x2.shape[0]
+
+    n = n_x1 + n_x2 - 1
+
+    fft1 = np.fft.rfft(x1, n=n)
+    fft2 = np.fft.rfft(x2, n=n)
+
+    if phat:
+        eps1 = np.mean(np.abs(fft1)) * 1e-10
+        fft1 /= np.abs(fft1) + eps1
+        eps2 = np.mean(np.abs(fft2)) * 1e-10
+        fft2 /= np.abs(fft2) + eps2
+
+    out = np.fft.irfft(fft1 * np.conj(fft2), n=int(n * interp))
+
+    return np.concatenate([out[-interp * (n_x2 - 1) :], out[: (interp * n_x1)]])

sonusai/mixture/mixdb.py

@@ -255,6 +255,16 @@ class MixtureDatabase:
                     "Predicted rating of overall quality of mixture versus true targets",
                 ),
                 MetricDoc("Mixture Metrics", "ssnr", "Segmental SNR"),
+                MetricDoc("Mixture Metrics", "mxdco", "Mixture DC offset"),
+                MetricDoc("Mixture Metrics", "mxmin", "Mixture min level"),
+                MetricDoc("Mixture Metrics", "mxmax", "Mixture max levl"),
+                MetricDoc("Mixture Metrics", "mxpkdb", "Mixture Pk lev dB"),
+                MetricDoc("Mixture Metrics", "mxlrms", "Mixture RMS lev dB"),
+                MetricDoc("Mixture Metrics", "mxpkr", "Mixture RMS Pk dB"),
+                MetricDoc("Mixture Metrics", "mxtr", "Mixture RMS Tr dB"),
+                MetricDoc("Mixture Metrics", "mxcr", "Mixture Crest factor"),
+                MetricDoc("Mixture Metrics", "mxfl", "Mixture Flat factor"),
+                MetricDoc("Mixture Metrics", "mxpkc", "Mixture Pk count"),
                 MetricDoc("Mixture Metrics", "mxtdco", "Mixture target DC offset"),
                 MetricDoc("Mixture Metrics", "mxtmin", "Mixture target min level"),
                 MetricDoc("Mixture Metrics", "mxtmax", "Mixture target max levl"),
@@ -681,7 +691,7 @@ class MixtureDatabase:
 
         with self.db() as c:
             return [
-                ImpulseResponseFile(impulse_response[1], json.loads(impulse_response[2]))
+                ImpulseResponseFile(impulse_response[1], json.loads(impulse_response[2]), impulse_response[3])
                 for impulse_response in c.execute(
                     "SELECT impulse_response_file.* FROM impulse_response_file"
                 ).fetchall()
@@ -700,15 +710,25 @@ class MixtureDatabase:
             ]
 
     def impulse_response_file(self, ir_id: int | None) -> str | None:
-        """Get impulse response file with ID from db
+        """Get impulse response file name with ID from db
 
         :param ir_id: Impulse response file ID
-        :return: Noise
+        :return: Impulse response file name
         """
         if ir_id is None:
             return None
         return _impulse_response_file(self.db, ir_id, self.use_cache)
 
+    def impulse_response_delay(self, ir_id: int | None) -> int | None:
+        """Get impulse response delay with ID from db
+
+        :param ir_id: Impulse response file ID
+        :return: Impulse response delay
+        """
+        if ir_id is None:
+            return None
+        return _impulse_response_delay(self.db, ir_id, self.use_cache)
+
     @cached_property
     def num_impulse_response_files(self) -> int:
         """Get number of impulse response files from db
@@ -814,18 +834,11 @@ class MixtureDatabase:
         :return: Augmented noise audio
         """
         from .audio import read_audio
-        from .audio import read_ir
         from .augmentation import apply_augmentation
-        from .augmentation import apply_impulse_response
 
         noise = self.noise_file(mixture.noise.file_id)
         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), self.use_cache),  # pyright: ignore [reportArgumentType]
-            )
+        audio = apply_augmentation(self, audio, mixture.noise.augmentation.pre)
 
         return audio
 
@@ -859,8 +872,9 @@ class MixtureDatabase:
         for target in mixture.targets:
             target_audio = self.read_target_audio(target.file_id)
             target_audio = apply_augmentation(
+                mixdb=self,
                 audio=target_audio,
-                augmentation=target.augmentation,
+                augmentation=target.augmentation.pre,
                 frame_length=self.feature_step_samples,
             )
             target_audio = apply_gain(audio=target_audio, gain=mixture.target_snr_gain)
@@ -1119,8 +1133,7 @@ class MixtureDatabase:
         offsets = range(0, mixture.samples, self.ft_config.overlap)
         if len(target_energy) != len(offsets):
             raise ValueError(
-                f"Number of frames in energy, {len(target_energy)},"
-                f" is not number of frames in mixture, {len(offsets)}"
+                f"Number of frames in energy, {len(target_energy)}, is not number of frames in mixture, {len(offsets)}"
             )
 
         for idx, offset in enumerate(offsets):
@@ -1370,11 +1383,11 @@ class MixtureDatabase:
                     # Check for tempo augmentation and adjust Interval start and end data as needed
                     entries = []
                     for entry in data:
-                        if target.augmentation.tempo is not None:
+                        if target.augmentation.pre.tempo is not None:
                             entries.append(
                                 Interval(
-                                    entry.start / target.augmentation.tempo,
-                                    entry.end / target.augmentation.tempo,
+                                    entry.start / target.augmentation.pre.tempo,
+                                    entry.end / target.augmentation.pre.tempo,
                                     entry.label,
                                 )
                             )
@@ -1595,6 +1608,19 @@ class MixtureDatabase:
 
         speech = create_speech()
 
+        def create_mixture_stats() -> Callable[[], AudioStatsMetrics]:
+            state: AudioStatsMetrics | None = None
+
+            def get() -> AudioStatsMetrics:
+                nonlocal state
+                if state is None:
+                    state = calc_audio_stats(mixture_audio(), self.fg_info.ft_config.length / SAMPLE_RATE)
+                return state
+
+            return get
+
+        mixture_stats = create_mixture_stats()
+
         def create_targets_stats() -> Callable[[], list[AudioStatsMetrics]]:
             state: list[AudioStatsMetrics] | None = None
 
@@ -1803,6 +1829,36 @@ class MixtureDatabase:
                     extended=False,
                 )
 
+            if m == "mxdco":
+                return mixture_stats().dco
+
+            if m == "mxmin":
+                return mixture_stats().min
+
+            if m == "mxmax":
+                return mixture_stats().max
+
+            if m == "mxpkdb":
+                return mixture_stats().pkdb
+
+            if m == "mxlrms":
+                return mixture_stats().lrms
+
+            if m == "mxpkr":
+                return mixture_stats().pkr
+
+            if m == "mxtr":
+                return mixture_stats().tr
+
+            if m == "mxcr":
+                return mixture_stats().cr
+
+            if m == "mxfl":
+                return mixture_stats().fl
+
+            if m == "mxpkc":
+                return mixture_stats().pkc
+
             if m == "mxtdco":
                 return target_stats().dco
 
@@ -2042,12 +2098,12 @@ def __noise_file(db: partial, n_id: int) -> NoiseFile:
 
 
 def _impulse_response_file(db: partial, ir_id: int, use_cache: bool = True) -> str:
-    """Get impulse response file with ID from db
+    """Get impulse response file name with ID from db
 
     :param db: Database context
     :param ir_id: Impulse response file ID
     :param use_cache: If true, use LRU caching
-    :return: Impulse response
+    :return: Impulse response file name
     """
     if use_cache:
         return __impulse_response_file(db, ir_id)
@@ -2069,6 +2125,34 @@ def __impulse_response_file(db: partial, ir_id: int) -> str:
         )
 
 
+def _impulse_response_delay(db: partial, ir_id: int, use_cache: bool = True) -> int:
+    """Get impulse response delay with ID from db
+
+    :param db: Database context
+    :param ir_id: Impulse response file ID
+    :param use_cache: If true, use LRU caching
+    :return: Impulse response delay
+    """
+    if use_cache:
+        return __impulse_response_delay(db, ir_id)
+    return __impulse_response_delay.__wrapped__(db, ir_id)
+
+
+@lru_cache
+def __impulse_response_delay(db: partial, ir_id: int) -> int:
+    with db() as c:
+        return int(
+            c.execute(
+                """
+                SELECT impulse_response_file.delay
+                FROM impulse_response_file
+                WHERE ? = impulse_response_file.id
+                """,
+                (ir_id + 1,),
+            ).fetchone()[0]
+        )
+
+
 def _mixture(db: partial, m_id: int, use_cache: bool = True) -> Mixture:
     """Get mixture record with ID from db
 

sonusai/mixture/targets.py

@@ -16,14 +16,11 @@ def get_augmented_targets(
 
     augmented_targets: list[AugmentedTarget] = []
     for mixup in mixups:
-        augmentation_indices = get_augmentation_indices_for_mixup(target_augmentations, mixup)
+        target_augmentation_indices = get_augmentation_indices_for_mixup(target_augmentations, mixup)
         for target_index in range(len(target_files)):
-            for augmentation_index in augmentation_indices:
+            for target_augmentation_index in target_augmentation_indices:
                 augmented_targets.append(
-                    AugmentedTarget(
-                        target_id=target_index,
-                        target_augmentation_id=augmentation_index,
-                    )
+                    AugmentedTarget(target_id=target_index, target_augmentation_id=target_augmentation_index)
                 )
 
     return augmented_targets

sonusai/utils/__init__.py

@@ -1,5 +1,6 @@
 # SonusAI general utilities
 # ruff: noqa: F401
+
 from .asl_p56 import asl_p56
 from .asr import ASRResult
 from .asr import calc_asr
@@ -53,5 +54,6 @@ from .stacked_complex import stacked_complex_imag
 from .stacked_complex import stacked_complex_real
 from .stacked_complex import unstack_complex
 from .stratified_shuffle_split import stratified_shuffle_split_mixid
+from .temp_seed import temp_seed
 from .write_audio import write_audio
 from .yes_or_no import yes_or_no

sonusai/utils/temp_seed.py

@@ -0,0 +1,13 @@
+import contextlib
+
+import numpy as np
+
+
+@contextlib.contextmanager
+def temp_seed(seed):
+    state = np.random.get_state()
+    np.random.seed(seed)
+    try:
+        yield
+    finally:
+        np.random.set_state(state)

{sonusai-0.19.10.dist-info → sonusai-0.20.2.dist-info}/METADATA

@@ -1,6 +1,6 @@
-Metadata-Version: 2.1
+Metadata-Version: 2.3
 Name: sonusai
-Version: 0.19.10
+Version: 0.20.2
 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-0.19.10.dist-info → sonusai-0.20.2.dist-info}/RECORD

@@ -3,7 +3,7 @@ sonusai/aawscd_probwrite.py,sha256=QZLMQrmPr3OjZ06buyYDwlnk9YPCpyr4KHkBjPsiqjU,3
 sonusai/audiofe.py,sha256=iFdthh4UrOvziT8urjrjD7dACWZPQz9orM5bVAW3WSQ,11269
 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/genmixdb.yml,sha256=qFK_VoUxHmc-EhJYZr4pAEY9tu6zRvloubq0NmrbH6I,956
 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
@@ -12,12 +12,13 @@ sonusai/deprecated/gentcst.py,sha256=nKbHy3aHreHqA-XnLQOzOApS8RuTNUFqnx52a8I5zLQ
 sonusai/deprecated/plot.py,sha256=xL0w8Dtjdns2KX8tbTrdBGXviy_aoV3WUJSVKPZkQng,17423
 sonusai/deprecated/tplot.py,sha256=0p238DvTaP4oU9y-dp0JdLaTV4TKrooAwbx7zdz_QAc,14641
 sonusai/doc/__init__.py,sha256=KyQ26Um0RM8A3GYsb_tbFH64RwpoAw6lja2f_moUWas,33
-sonusai/doc/doc.py,sha256=VZXauwbOb-VIufWw-lu0yfrd6jMRPeFeVPaaEjZNvn4,18881
+sonusai/doc/doc.py,sha256=nEnvau0PIl2xbxET8AjFkwTLVShemmQ4CiTio94aLg0,19275
 sonusai/doc.py,sha256=zSmXpioB0YS_5-7kqfS5cr--veSaXkxRKzldId9Hyoc,878
 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/ir_metric.py,sha256=n35_RssAk2jjqm1iXJ6euMtK00LV4qohdBfDAZZpNlU,19581
 sonusai/lsdb.py,sha256=0HOGDDndB3LT9cz9AaxKIpt9vslAoSP4F239gply4Xg,5149
 sonusai/main.py,sha256=HbnEia1B1-Z-mlHkLfojH8aj9GIpL1Btw3oH60T_CCQ,2590
 sonusai/metrics/__init__.py,sha256=ssV6JEK_oklRSocsp6HMcG-GtJvV8IkRQtdKhHHmwU8,878
@@ -37,31 +38,27 @@ 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/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/metrics_summary.py,sha256=DchpgBNYcBPz4t1YRindCm1CVmJLmXY7-oyaXpxBnWA,12106
+sonusai/mixture/__init__.py,sha256=9TE21nlj4TOrSPopLh5Lh769v68v0kWgNkr_GGk9TEs,5300
+sonusai/mixture/audio.py,sha256=MlsuhY8Zc8puBClO5utBP3mveAk8MLZtXvF6ztuisW4,6629
+sonusai/mixture/augmentation.py,sha256=UYNdfVcYCsmegicKpGLTsYyvO97NQOTLoZjwI9sZROk,14646
 sonusai/mixture/class_count.py,sha256=zcC3BDYMPN6wJYmO1RcOuqmrnTQIbMSznl33oN3e2sc,597
-sonusai/mixture/config.py,sha256=g5ZmOhFYqmEdRQYSgfDIZ9VM0QiTwBqk7vIyAvxnPMo,24211
-sonusai/mixture/constants.py,sha256=fXcWuSI4YZOAuncLGEUeEW9WWNZeN-6mI8LFNILwyTc,1494
+sonusai/mixture/config.py,sha256=gZUPLGl7VKW32HfnFMEajKgJE35ZJ_edpxaA0TRPE1E,24469
+sonusai/mixture/constants.py,sha256=yGXNjB87boJVSt1Q8hBTrNzOP0XVZcFf1k34u1yyUWU,1481
 sonusai/mixture/data_io.py,sha256=KZGqhHd9_ucAfZEAXPIc5XL3aHYgdV5CyqaBx5_t8OM,5551
-sonusai/mixture/datatypes.py,sha256=xNDBWFTVQ3plJ7qHKzrXyV4pffPYuf1xMVqBsR40n4o,10487
+sonusai/mixture/datatypes.py,sha256=LFWDsgGVY7Z3VPjWaB5g8q5Ss8v5bR2EfU6ygI0wZaU,10707
 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=yoJOcY9KPe_B1RVnENVr4ekcnXyZJMdvKMbJggpLOi4,38084
-sonusai/mixture/helpers.py,sha256=Bt9njNb_OZ3j02qgrVEMZiL0hX4kXtFK_tkPoGoeb4Y,15787
+sonusai/mixture/generation.py,sha256=f3DsDvVAAdpvh1lBWV-jMj5qNeZ2qmR9RA_4eI4NDcU,37954
+sonusai/mixture/helpers.py,sha256=nNqK__MBp8f10telUU3A8FhkNeXYPGCx8dxxDpzCAbg,15464
+sonusai/mixture/ir_delay.py,sha256=WRoYjuHpsppe0D5qQExNxsCyKbMPDfWJ4CTlr3Ps50k,2036
 sonusai/mixture/log_duration_and_sizes.py,sha256=qhgl87C2KbjxLdKEpjYOoqNL6rc-8-PB4R7Gx_7UG8g,1240
-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
+sonusai/mixture/mixdb.py,sha256=oMBbi0HRiEBtN7lDup8qhPBIQ_td89CHa0bYy2PXdWQ,78744
 sonusai/mixture/spectral_mask.py,sha256=U9XJ_SAoI9b67K_3SE7bNw6U8cPGFOBttaZAxMjA_Jc,2042
 sonusai/mixture/target_class_balancing.py,sha256=o_TZ8kVYq10lgeXHh3GUFfflfdUvRt0FekFu2eaNkDs,4251
-sonusai/mixture/targets.py,sha256=6emo2fxxp9ZhSpHuUM9xIjYMz8zeIHAw684jT3l7fAs,6442
+sonusai/mixture/targets.py,sha256=oOeqdE-n-sCq_9luEt82HEP0MRCaHG_7J-p3nCftkAc,6399
 sonusai/mixture/tokenized_shell_vars.py,sha256=lXTzUDutuBWGV1zIsqeIxWmy-eKm0Vx1y8-iLdsL1gQ,4921
-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
@@ -86,7 +83,7 @@ sonusai/speech/types.py,sha256=4eKVPAktpkIrZ2qoVp2iT45zxTVNocQEGT6O_Zlub_w,214
 sonusai/speech/vctk.py,sha256=WInvRRRkZCW6t_NcZAJffJzgCbyetal-j2w0kKX5SDw,1527
 sonusai/speech/voxceleb.py,sha256=Uu1kB1krf8hess1yuvGbYfV_VgYhklEyoz4I7KfrVpw,2658
 sonusai/summarize_metric_spenh.py,sha256=2w81ZgJahYvD6wCpE3DFoUFrXexLXjO44ITRVm1HJXw,1858
-sonusai/utils/__init__.py,sha256=z72OlzZCHpYfYHKnHn7jznj6Zt7zB-FyO6hIgFk45As,2379
+sonusai/utils/__init__.py,sha256=D7IFq4Ozy_DRq4pC50YRxGQybqWEDXglCuAgxNIpFyU,2413
 sonusai/utils/asl_p56.py,sha256=cPUVwXawF7vLJgs4zUtoRGk7Wdbe5KKti_-v_8xIU10,3862
 sonusai/utils/asr.py,sha256=ubiU3E61HN3r9MhPV7ci37cnLZowll8KfjUS7os3Sho,2822
 sonusai/utils/asr_functions/__init__.py,sha256=HKGRm_c48tcxlfwqH63m-MvhAoK_pCcw76lxmFmiP_U,63
@@ -121,10 +118,11 @@ 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=fcGW8nkZIwUqq1qtxbK_ZH58sYULqZfv7iNBQnKGH-M,6706
+sonusai/utils/temp_seed.py,sha256=Ava5TCGpvDBtaRx2l-40CuGIjhgLevu1KFfZsgr38qM,218
 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-0.19.10.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,,
+sonusai-0.20.2.dist-info/METADATA,sha256=CT_z1tJfku142nH0PL22DT7EEc8bFfJj9qyWcR7v6FU,2535
+sonusai-0.20.2.dist-info/WHEEL,sha256=RaoafKOydTQ7I_I3JTrPCg6kUmTgtm4BornzOqyEfJ8,88
+sonusai-0.20.2.dist-info/entry_points.txt,sha256=zMNjEphEPO6B3cD1GNpit7z-yA9tUU5-j3W2v-UWstU,92
+sonusai-0.20.2.dist-info/RECORD,,

{sonusai-0.19.10.dist-info → sonusai-0.20.2.dist-info}/WHEEL

@@ -1,4 +1,4 @@
 Wheel-Version: 1.0
-Generator: poetry-core 1.9.1
+Generator: poetry-core 2.0.0
 Root-Is-Purelib: true
 Tag: py3-none-any