sonusai 0.15.6__py3-none-any.whl → 0.15.8__py3-none-any.whl

sonusai/mixture/generation.py

@@ -433,7 +433,7 @@ def _initialize_targets_audio(mixdb: MixtureDatabase, mixture: Mixture) -> tuple
         # 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).
         if target.augmentation.gain is not None:
-            target.gain = round(10 ** (target.augmentation.gain / 20), ndigits=7)
+            target.gain = round(10 ** (target.augmentation.gain / 20), ndigits=5)
         else:
             target.gain = 1
 
@@ -507,8 +507,8 @@ def _initialize_mixture_gains(mixdb: MixtureDatabase,
         mixture.target_snr_gain *= gain_adjustment
         mixture.noise_snr_gain *= gain_adjustment
 
-    mixture.target_snr_gain = round(mixture.target_snr_gain, ndigits=7)
-    mixture.noise_snr_gain = round(mixture.noise_snr_gain, ndigits=7)
+    mixture.target_snr_gain = round(mixture.target_snr_gain, ndigits=5)
+    mixture.noise_snr_gain = round(mixture.noise_snr_gain, ndigits=5)
     return mixture
 
 

sonusai/mixture/truth_functions/data.py

@@ -23,6 +23,8 @@ class Data:
                               num_classes=config.num_classes,
                               truth_mutex=config.mutex)
 
+        self.num_bands = fg.num_bands
+        self.ttype = fg.ftransform_ttype
         self.frame_size = fg.ftransform_R
 
         if len(target_audio) % self.frame_size != 0:

sonusai/mixture/truth_functions/target.py

@@ -13,12 +13,13 @@ Calculates the true transform of the target using the STFT
 configuration defined by the feature. This will include a
 forward transform window if defined by the feature.
 
-Output shape: [:, 2 * bins] (stacked real, imag)
+Output shape: [:, num_classes]
+                    (target stacked real, imag; or real only for tdac-co)
     """
 
     from sonusai import SonusAIError
 
-    if data.config.num_classes != 2 * data.target_fft.bins:
+    if data.config.num_classes != data.num_bands:
         raise SonusAIError(f'Invalid num_classes for target_f truth: {data.config.num_classes}')
 
     target_freq = _execute_fft(data.target_audio, data.target_fft, len(data.offsets))
@@ -28,6 +29,7 @@ Output shape: [:, 2 * bins] (stacked real, imag)
                                       frame_size=data.frame_size,
                                       zero_based_indices=data.zero_based_indices,
                                       bins=data.target_fft.bins,
+                                      ttype=data.ttype,
                                       start=0,
                                       truth=data.truth)
 
@@ -43,11 +45,13 @@ using the STFT configuration defined by the feature. This
 will include a forward transform window if defined by the
 feature.
 
-Output shape: [:, 4 * bins] (target stacked real, imag; mixture stacked real, imag)
+Output shape: [:, 2 * num_classes]
+                    (target stacked real, imag; or real only for tdac-co)
+                    (mixture stacked real, imag; or real only for tdac-co)
     """
     from sonusai import SonusAIError
 
-    if data.config.num_classes != 2 * data.target_fft.bins + 2 * data.mixture_fft.bins:
+    if data.config.num_classes != 2 * data.num_bands:
         raise SonusAIError(f'Invalid num_classes for target_mixture_f truth: {data.config.num_classes}')
 
     target_freq = _execute_fft(data.target_audio, data.target_fft, len(data.offsets))
@@ -59,6 +63,7 @@ Output shape: [:, 4 * bins] (target stacked real, imag; mixture stacked real, im
                                       frame_size=data.frame_size,
                                       zero_based_indices=data.zero_based_indices,
                                       bins=data.target_fft.bins,
+                                      ttype=data.ttype,
                                       start=0,
                                       truth=data.truth)
 
@@ -67,6 +72,7 @@ Output shape: [:, 4 * bins] (target stacked real, imag; mixture stacked real, im
                                       frame_size=data.frame_size,
                                       zero_based_indices=data.zero_based_indices,
                                       bins=data.target_fft.bins,
+                                      ttype=data.ttype,
                                       start=data.target_fft.bins * 2,
                                       truth=data.truth)
 
@@ -125,6 +131,7 @@ def _stack_real_imag(data: AudioF,
                      frame_size: int,
                      zero_based_indices: list[int],
                      bins: int,
+                     ttype: str,
                      start: int,
                      truth: Truth) -> Truth:
     import numpy as np
@@ -134,7 +141,8 @@ def _stack_real_imag(data: AudioF,
         b = _get_bin_slice(index + start, bins)
         truth[i, b] = np.real(data)
 
-        b = _get_bin_slice(b.stop, bins)
-        truth[i, b] = np.imag(data)
+        if ttype != 'tdac-co':
+            b = _get_bin_slice(b.stop, bins)
+            truth[i, b] = np.imag(data)
 
     return truth

{sonusai-0.15.6.dist-info → sonusai-0.15.8.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: sonusai
-Version: 0.15.6
+Version: 0.15.8
 Summary: Framework for building deep neural network models for sound, speech, and voice AI
 Home-page: https://aaware.com
 License: GPL-3.0-only
@@ -16,21 +16,21 @@ Classifier: Programming Language :: Python :: 3.10
 Classifier: Programming Language :: Python :: 3.11
 Requires-Dist: PyYAML (>=6.0.1,<7.0.0)
 Requires-Dist: aixplain (>=0.2.6,<0.3.0)
+Requires-Dist: ctranslate2 (==4.1.0)
 Requires-Dist: dataclasses-json (>=0.6.1,<0.7.0)
 Requires-Dist: deepgram-sdk (>=3.0.0,<4.0.0)
 Requires-Dist: docopt (>=0.6.2,<0.7.0)
-Requires-Dist: faster-whisper (>=0.10.0,<0.11.0)
-Requires-Dist: ffmpeg (>=1.4,<2.0)
-Requires-Dist: greenlet (>=3.0.1,<4.0.0)
-Requires-Dist: grpcio (==1.60.0)
+Requires-Dist: faster-whisper (>=1.0.1,<2.0.0)
+Requires-Dist: h5py (>=3.11.0,<4.0.0)
 Requires-Dist: jiwer (>=3.0.3,<4.0.0)
-Requires-Dist: keras-tuner (>=1.4.5,<2.0.0)
+Requires-Dist: keras (>=3.1.1,<4.0.0)
+Requires-Dist: keras-tuner (>=1.4.7,<2.0.0)
 Requires-Dist: librosa (>=0.10.1,<0.11.0)
-Requires-Dist: lightning (>=2.1.0,<3.0.0)
+Requires-Dist: lightning (>=2.2,<2.3)
 Requires-Dist: matplotlib (>=3.8.0,<4.0.0)
-Requires-Dist: onnx (==1.14.1)
+Requires-Dist: onnx (>=1.14.1,<2.0.0)
 Requires-Dist: onnxruntime (>=1.16.1,<2.0.0)
-Requires-Dist: paho-mqtt (>=1.6.1,<2.0.0)
+Requires-Dist: paho-mqtt (>=2.0.0,<3.0.0)
 Requires-Dist: pandas (>=2.1.1,<3.0.0)
 Requires-Dist: pesq (>=0.0.4,<0.0.5)
 Requires-Dist: pyaaware (>=1.5.3,<2.0.0)
@@ -38,18 +38,14 @@ Requires-Dist: pydub (>=0.25.1,<0.26.0)
 Requires-Dist: pystoi (>=0.4.0,<0.5.0)
 Requires-Dist: python-magic (>=0.4.27,<0.5.0)
 Requires-Dist: requests (>=2.31.0,<3.0.0)
-Requires-Dist: resampy (>=0.4.2,<0.5.0)
-Requires-Dist: samplerate (>=0.1.0,<0.2.0)
-Requires-Dist: scikit-learn (>=1.3.1,<2.0.0)
-Requires-Dist: sh (>=2.0.6,<3.0.0)
+Requires-Dist: samplerate (>=0.2.1,<0.3.0)
 Requires-Dist: soundfile (>=0.12.1,<0.13.0)
 Requires-Dist: sox (>=1.4.1,<2.0.0)
-Requires-Dist: speechrecognition (>=3.10.0,<4.0.0)
+Requires-Dist: speechrecognition (>=3.10.1,<4.0.0)
 Requires-Dist: tensorflow (>=2.15.0,<3.0.0)
-Requires-Dist: tensorflow-addons (>=0.23.0,<0.24.0)
 Requires-Dist: tf2onnx (>=1.15.1,<2.0.0)
-Requires-Dist: torch (>=2.1.0,<3.0.0)
-Requires-Dist: torchaudio (>=2.1.0,<3.0.0)
+Requires-Dist: torch (>=2.2,<2.3)
+Requires-Dist: torchaudio (>=2.2,<2.3)
 Requires-Dist: torchinfo (>=1.8.0,<2.0.0)
 Requires-Dist: tqdm (>=4.66.1,<5.0.0)
 Description-Content-Type: text/x-rst
@@ -59,7 +55,7 @@ Sonus AI: Framework for simplified creation of deep NN models for sound, speech,
 Sonus AI includes functions for pre-processing training and validation data and
 creating performance metrics reports for key types of Keras models:
 - recurrent, convolutional, or a combination (i.e. RCNNs)
-- binary, multiclass single-label, multiclass multi-label, and regresssion
+- binary, multiclass single-label, multiclass multi-label, and regression
 - training with data augmentations:  noise mixing, pitch and time stretch, etc.
 
 Sonus AI python functions are used by:

{sonusai-0.15.6.dist-info → sonusai-0.15.8.dist-info}/RECORD

@@ -44,7 +44,7 @@ sonusai/mixture/constants.py,sha256=xjCskcQi6khqYZDf7j6z1OkeN1C6wE06kBBapcJiNI4,
 sonusai/mixture/datatypes.py,sha256=xN-GdPCEHGE2Ak_TdFbjuSyMs4x7TLRp59trbMTiYLg,8164
 sonusai/mixture/eq_rule_is_valid.py,sha256=MpQwRA5M76wSiQWEI1lW2cLFdPaMttBLcQp3tWD8efM,1243
 sonusai/mixture/feature.py,sha256=io6OiJAJ3GYvPChiUmPQuP3h0OB2onjYF8o9-AWkmqM,1996
-sonusai/mixture/generation.py,sha256=5SXV1R9os4XPN0pVm8ribG23vsXkYzsegoctRZdXQT4,39039
+sonusai/mixture/generation.py,sha256=miUrc3QOSUNIG6mDkiMCZ6M2ulivUZxlYUAJUOVomWc,39039
 sonusai/mixture/helpers.py,sha256=XqpcB15MezEMVJwf3jxzATDJSpj_27b8Cru1TDIFD7w,21326
 sonusai/mixture/log_duration_and_sizes.py,sha256=r-wVjrLW1XBciOL4pkZSYMR7ZNADbojE95TPSQkp3kc,1329
 sonusai/mixture/mapped_snr_f.py,sha256=mlbYM1t14OXe_Zg4CjpWTuA_Zun4W0O3bSUXeodRBQs,1845
@@ -61,12 +61,12 @@ sonusai/mixture/torchaudio_augmentation.py,sha256=1vEDHI0caL1vrgoY2lAWe4CiHE2jKR
 sonusai/mixture/truth.py,sha256=Y41pZ52Xkols9LUler0NlgnilUOscBIucmw4GcxXNzU,1612
 sonusai/mixture/truth_functions/__init__.py,sha256=82lKYHhLy8KW3gHngrocoqwupGVLVsWdIXdYs3vhjOc,359
 sonusai/mixture/truth_functions/crm.py,sha256=_Vy8UMrOUQXsrM3nutvUMWCpvI8GePr01QFlyqLFd4k,2626
-sonusai/mixture/truth_functions/data.py,sha256=wUlZAAeE4AXg4Egtcvs7Qhbf5dnWur-1Qr_Dz6GoJiQ,2778
+sonusai/mixture/truth_functions/data.py,sha256=NJNZz5fB3jnntUDlnsKJVQIeuHNUvD4x5iNaQVQlo3Y,2857
 sonusai/mixture/truth_functions/energy.py,sha256=ydMtMLjMloG76DB30ZHQ5tkBVh4dkMJ82XEhKBokmIk,4281
 sonusai/mixture/truth_functions/file.py,sha256=jOJuC_3y9BH6GGOp9eKcbVrHLVRzUA80BJq59LhcBUM,1539
 sonusai/mixture/truth_functions/phoneme.py,sha256=stYdlPuNytQK_LLT61OJLfYSqKd-sDjQZdtJKGzt5wA,479
 sonusai/mixture/truth_functions/sed.py,sha256=8cHjEFjZaH_0hIOHhPmj4AJz2GpEADM6Ys2x4NoiWSY,2469
-sonusai/mixture/truth_functions/target.py,sha256=V-bXcJZ3P83NHKqYs4RcsxxuZdO8kqzSEN5CNRa5e0k,5394
+sonusai/mixture/truth_functions/target.py,sha256=3rPXYwU4SBiPP3uIDpOL-B2Xw1Zh3JboD_MYNEyUpuk,5746
 sonusai/mkmanifest.py,sha256=dIPVFKKhnhHdq63OGr6p__pK7fyx3OdKVtbmGUJxsR8,7078
 sonusai/mkwav.py,sha256=LZNyhq4gJEs_NtGvRsYHA2qfgkkODpt6HoH1b-Tjjuw,5266
 sonusai/onnx_predict.py,sha256=RhQbbNG3w6rCXuSFUWCaQmUH5JzSP2hmu6TG5_81IVA,9055
@@ -122,7 +122,7 @@ sonusai/utils/trim_docstring.py,sha256=dSrtiRsEN4wkkvKBp6WDr13RUypfqZzgH_jOBLs1o
 sonusai/utils/wave.py,sha256=TKE-CNPGFXNXUW626CBPzCTNgWJut8I0ZEUsgG9q4Po,586
 sonusai/utils/yes_or_no.py,sha256=eMLXBVH0cEahiXY4W2KNORmwNQ-ba10eRtldh0y4NYg,263
 sonusai/vars.py,sha256=m2AefF0m5bXWGXpJj8Pi42zWL2ydeEj7bkak3GrtMyM,940
-sonusai-0.15.6.dist-info/METADATA,sha256=bUbyCYkQJaHaxBfqMyieWqqGRewQJ3Vu0LFzGpOR3Lg,3096
-sonusai-0.15.6.dist-info/WHEEL,sha256=FMvqSimYX_P7y0a7UY-_Mc83r5zkBZsCYPm7Lr0Bsq4,88
-sonusai-0.15.6.dist-info/entry_points.txt,sha256=zMNjEphEPO6B3cD1GNpit7z-yA9tUU5-j3W2v-UWstU,92
-sonusai-0.15.6.dist-info/RECORD,,
+sonusai-0.15.8.dist-info/METADATA,sha256=3eCpCJmXOfr7GV3a7HDWo0iilEVHB5ANdQqS59O0Yi0,2920
+sonusai-0.15.8.dist-info/WHEEL,sha256=FMvqSimYX_P7y0a7UY-_Mc83r5zkBZsCYPm7Lr0Bsq4,88
+sonusai-0.15.8.dist-info/entry_points.txt,sha256=zMNjEphEPO6B3cD1GNpit7z-yA9tUU5-j3W2v-UWstU,92
+sonusai-0.15.8.dist-info/RECORD,,