sonusai 0.18.2__py3-none-any.whl → 0.18.4__py3-none-any.whl

sonusai/mixture/truth_functions/energy.py

@@ -1,12 +1,11 @@
+import numpy as np
+
 from sonusai.mixture.datatypes import Truth
 from sonusai.mixture.truth_functions.data import Data
 
 
 def _core(data: Data, mapped: bool, snr: bool) -> Truth:
-    import numpy as np
-
     from sonusai import SonusAIError
-    from sonusai.mixture import calculate_mapped_snr_f
     from sonusai.utils import compute_energy_f
 
     snr_db_mean = None
@@ -52,7 +51,7 @@ def _core(data: Data, mapped: bool, snr: bool) -> Truth:
         tmp = np.nan_to_num(tmp, nan=-np.inf, posinf=np.inf, neginf=-np.inf)
 
         if mapped:
-            tmp = calculate_mapped_snr_f(tmp, snr_db_mean, snr_db_std)
+            tmp = _calculate_mapped_snr_f(tmp, snr_db_mean, snr_db_std)
 
         for index in data.zero_based_indices:
             data.truth[offset:offset + data.frame_size, index:index + data.target_fft.bins] = tmp
@@ -60,16 +59,31 @@ def _core(data: Data, mapped: bool, snr: bool) -> Truth:
     return data.truth
 
 
+def _calculate_mapped_snr_f(truth_f: np.ndarray, snr_db_mean: np.ndarray, snr_db_std: np.ndarray) -> np.ndarray:
+    """Calculate mapped SNR from standard SNR energy per bin/class."""
+    import scipy.special as sc
+
+    old_err = np.seterr(divide='ignore', invalid='ignore')
+    num = 10 * np.log10(np.double(truth_f)) - np.double(snr_db_mean)
+    den = np.double(snr_db_std) * np.sqrt(2)
+    q = num / den
+    q = np.nan_to_num(q, nan=-np.inf, posinf=np.inf, neginf=-np.inf)
+    result = 0.5 * (1 + sc.erf(q))
+    np.seterr(**old_err)
+
+    return result.astype(np.float32)
+
+
 def energy_f(data: Data) -> Truth:
     """Frequency domain energy truth generation function
 
-Calculates the true energy per bin:
+    Calculates the true energy per bin:
 
-Ti^2 + Tr^2
+    Ti^2 + Tr^2
 
-where T is the target STFT bin values.
+    where T is the target STFT bin values.
 
-Output shape: [:, bins]
+    Output shape: [:, bins]
     """
     return _core(data=data, mapped=False, snr=False)
 
@@ -77,13 +91,13 @@ Output shape: [:, bins]
 def snr_f(data: Data) -> Truth:
     """Frequency domain SNR truth function documentation
 
-Calculates the true SNR per bin:
+    Calculates the true SNR per bin:
 
-(Ti^2 + Tr^2) / (Ni^2 + Nr^2)
+    (Ti^2 + Tr^2) / (Ni^2 + Nr^2)
 
-where T is the target and N is the noise STFT bin values.
+    where T is the target and N is the noise STFT bin values.
 
-Output shape: [:, bins]
+    Output shape: [:, bins]
     """
     return _core(data=data, mapped=False, snr=True)
 
@@ -91,7 +105,7 @@ Output shape: [:, bins]
 def mapped_snr_f(data: Data) -> Truth:
     """Frequency domain mapped SNR truth function documentation
 
-Output shape: [:, bins]
+    Output shape: [:, bins]
     """
     return _core(data=data, mapped=True, snr=True)
 
@@ -99,27 +113,25 @@ Output shape: [:, bins]
 def energy_t(data: Data) -> Truth:
     """Time domain energy truth function documentation
 
-Calculates the true time domain energy of each frame:
+    Calculates the true time domain energy of each frame:
 
-For OLS:
-    sum(x[0:N-1]^2) / N
+    For OLS:
+        sum(x[0:N-1]^2) / N
 
-For OLA:
-    sum(x[0:R-1]^2) / R
+    For OLA:
+        sum(x[0:R-1]^2) / R
 
-where x is the target time domain data,
-N is the size of the transform, and
-R is the number of new samples in the frame.
+    where x is the target time domain data,
+    N is the size of the transform, and
+    R is the number of new samples in the frame.
 
-Output shape: [:, 1]
+    Output shape: [:, 1]
 
-Note: feature transforms can be defined to use a subset of all bins,
-i.e., subset of 0:128 for N=256 could be 0:127 or 1:128. energy_t
-will reflect the total energy over all bins regardless of the feature
-transform config.
+    Note: feature transforms can be defined to use a subset of all bins,
+    i.e., subset of 0:128 for N=256 could be 0:127 or 1:128. energy_t
+    will reflect the total energy over all bins regardless of the feature
+    transform config.
     """
-    import numpy as np
-
     from sonusai import SonusAIError
 
     _, target_energy = data.target_fft.execute_all(data.target_audio)

sonusai/mixture/truth_functions/target.py

@@ -16,7 +16,6 @@ forward transform window if defined by the feature.
 Output shape: [:, num_classes]
                     (target stacked real, imag; or real only for tdac-co)
     """
-
     from sonusai import SonusAIError
 
     if data.config.num_classes != data.feature_parameters:

sonusai/utils/__init__.py

@@ -1,8 +1,8 @@
 # SonusAI general utilities
 from .asl_p56 import asl_p56
-from .asr import ASRData
 from .asr import ASRResult
 from .asr import calc_asr
+from .asr import get_available_engines
 from .audio_devices import get_default_input_device
 from .audio_devices import get_input_device_index_by_name
 from .audio_devices import get_input_devices

sonusai/utils/asr.py

@@ -1,22 +1,10 @@
 from dataclasses import dataclass
-from typing import Any
 from typing import Callable
 from typing import Optional
 
 from sonusai.mixture import AudioT
 
 
-@dataclass(frozen=True)
-class ASRData:
-    audio: AudioT
-    whisper_model: Optional[Any] = None
-    whisper_model_name: Optional[str] = None
-    device: Optional[str] = None
-    cpu_threads: Optional[int] = None
-    compute_type: Optional[str] = None
-    beam_size: Optional[int] = None
-
-
 @dataclass(frozen=True)
 class ASRResult:
     text: str
@@ -28,57 +16,56 @@ class ASRResult:
     asr_cpu_time: Optional[float] = None
 
 
-def get_available_engines() -> dict[str, Callable[[ASRData], ASRResult]]:
+def get_available_engines() -> list[str]:
     from importlib import import_module
     from pkgutil import iter_modules
 
     module = import_module('sonusai.utils.asr_functions')
-    engines = {method: getattr(module, method) for method in dir(module) if not method.startswith('_')}
+    engines = [method for method in dir(module) if not method.startswith('_')]
     for _, name, _ in iter_modules():
         if name.startswith('sonusai_asr_'):
             module = import_module(f'{name}.asr_functions')
             for method in dir(module):
                 if not method.startswith('_'):
-                    engines[method] = getattr(module, method)
+                    engines.append(method)
 
     return engines
 
 
-def calc_asr(audio: AudioT | str,
-             engine: Optional[str] = 'aaware_whisper',
-             whisper_model: Optional[Any] = None,
-             whisper_model_name: Optional[str] = 'tiny',
-             device: Optional[str] = 'cpu',
-             cpu_threads: Optional[int] = 1,
-             compute_type: Optional[str] = 'int8',
-             beam_size: Optional[int] = 5) -> ASRResult:
+def _asr_fn(engine: str) -> Callable[..., ASRResult]:
+    from importlib import import_module
+    from pkgutil import iter_modules
+
+    module = import_module('sonusai.utils.asr_functions')
+    for method in dir(module):
+        if method == engine:
+            return getattr(module, method)
+
+    for _, name, _ in iter_modules():
+        if name.startswith('sonusai_asr_'):
+            module = import_module(f'{name}.asr_functions')
+            for method in dir(module):
+                if method == engine:
+                    return getattr(module, method)
+
+    raise ValueError(f'engine {engine} not supported')
+
+
+def calc_asr(audio: AudioT | str, engine: str, **config) -> ASRResult:
     """Run ASR on audio
 
     :param audio: Numpy array of audio samples or location of an audio file
-    :param engine: Type of ASR engine to use
-    :param whisper_model: A preloaded Whisper ASR model
-    :param whisper_model_name: Name of Whisper ASR model to use if none was provided
-    :param device: the device to put the ASR model into
-    :param cpu_threads: int specifying threads to use when device is cpu
-           note: must be 1 if this func is run in parallel
-    :param compute_type: the precision of ASR model to use
-    :param beam_size: int specifying beam_size to use
+    :param engine: ASR engine to use
+    :param config: kwargs configuration parameters
     :return: ASRResult object containing text and confidence
     """
     from copy import copy
 
     import numpy as np
 
-    from sonusai import SonusAIError
     from sonusai.mixture import read_audio
 
-    available_engines = get_available_engines()
-    if engine not in available_engines:
-        raise SonusAIError(f'Unsupported ASR function: {engine}')
-
     if not isinstance(audio, np.ndarray):
         audio = copy(read_audio(audio))
 
-    data = ASRData(audio, whisper_model, whisper_model_name, device, cpu_threads, compute_type, beam_size)
-
-    return available_engines[engine](data)
+    return _asr_fn(engine)(audio, **config)

sonusai/utils/asr_functions/aaware_whisper.py

@@ -1,8 +1,8 @@
-from sonusai.utils import ASRData
+from sonusai.mixture import AudioT
 from sonusai.utils import ASRResult
 
 
-def aaware_whisper(data: ASRData) -> ASRResult:
+def aaware_whisper(audio: AudioT, **_config) -> ASRResult:
     import tempfile
     from math import exp
     from os import getenv
@@ -22,7 +22,7 @@ def aaware_whisper(data: ASRData) -> ASRResult:
 
     with tempfile.TemporaryDirectory() as tmp:
         file = join(tmp, 'asr.wav')
-        write_audio(name=file, audio=float_to_int16(data.audio))
+        write_audio(name=file, audio=float_to_int16(audio))
 
         files = {'audio_file': (file, open(file, 'rb'), 'audio/wav')}
 

{sonusai-0.18.2.dist-info → sonusai-0.18.4.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: sonusai
-Version: 0.18.2
+Version: 0.18.4
 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-0.18.2.dist-info → sonusai-0.18.4.dist-info}/RECORD

@@ -1,67 +1,70 @@
-sonusai/__init__.py,sha256=j2eH_QUsIIMm0HDiNHC5HCocWsX_GhtvlmTkT7zyYOw,2918
+sonusai/__init__.py,sha256=s77La_fCT4TJtXR8fT0jHrWgGwOQI5qonyS3LStzsyw,2980
 sonusai/aawscd_probwrite.py,sha256=GukR5owp_0A3DrqSl9fHWULYgclNft4D5OkHIwfxxkc,3698
-sonusai/audiofe.py,sha256=zOySiYs5ZZm60eMbA7RjhG6C0Ouhaii3WfL1d0Q8rxg,11154
-sonusai/calc_metric_spenh.py,sha256=SunJD8wkdUxyL0rRZt2auauZBEUzpi0IRY8MtXKh3wo,63645
+sonusai/audiofe.py,sha256=3LssRiL73DH8teihD9f3nCvfZ0a65WQtXCqWGnKHuJM,11157
+sonusai/calc_metric_spenh.py,sha256=ee2xrx6L1lFyWSoQSiq56He3RQ1cF7T_ak-6TjejXsc,47738
 sonusai/data/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-sonusai/data/genmixdb.yml,sha256=-XSs_hUR6wHJVoTPmSewzXL7u61X-xmHY46lNPatxSE,1025
+sonusai/data/genmixdb.yml,sha256=p7AhJjtQ4S_xtDhVsFEqUi8fmvNzCLCV1LPdoNYvmWo,1043
 sonusai/data/speech_ma01_01.wav,sha256=PK0vMKg-NR6rPE3KouxHGF6PKXnJCr7AwjMqfu98LUA,76644
 sonusai/data/whitenoise.wav,sha256=I2umov0m34y56F9IsIBi1XtE76ZeZaSKDf70cJRe3pI,1920044
 sonusai/doc/__init__.py,sha256=rP5Hgn0Iys_xkuv4caxngdqehuU4zLZsiKuv8Nde67M,19
-sonusai/doc/doc.py,sha256=LOf4HiIxMeQ7-n0ExDSldo1Tt1036SVaN_auwlqXUxQ,21489
+sonusai/doc/doc.py,sha256=8gftMlCaVa0hIA_BjSHbEYgSTq_BAZwC5Ipk-BFgGp8,22919
 sonusai/doc.py,sha256=l8CaFgLI8mqx4tn0aXfxKqa2dy9GgC0zjYxZAkpmi1E,878
 sonusai/genft.py,sha256=OzET3iTE-QhrUckzidfZvCDXZlAxIF5Xe5NEf856Vvk,5662
-sonusai/genmetrics.py,sha256=fC8KPIB6wtBj_hs1X65lq3dqNTcWeuNs1eT7yXdpxD8,1830
+sonusai/genmetrics.py,sha256=YS5ZuhOR77-wGH8ju7z_ihQSDhYhV_iJWlWWMemLIGU,4998
 sonusai/genmix.py,sha256=TU5aTebGHsbfwsRbynYbegGBelSma9khuQkDk0dFE3I,7075
 sonusai/genmixdb.py,sha256=gF2qas1tH9MHEFLoEMrN3kYVm-vhAKaOuZ8ev-w4MQM,19553
 sonusai/gentcst.py,sha256=W1ZO3xs7CoZkFcvOTH-FLJOIA4I7Wzb0HVRC3hGGSaM,20223
-sonusai/lsdb.py,sha256=fMRqPlAu4B-4MsTXX-NaWXYyJ_dAOJlS-LrvQPQQsXg,6028
+sonusai/lsdb.py,sha256=oLyGoq46hkravDUXhI-wwhKdX_GESoOD5-E7nEJl7rc,5996
 sonusai/main.py,sha256=GC-pQrSqx9tWwIcmEo6V9SraEv5KskBLS_W_wz-f2ZM,2509
-sonusai/metrics/__init__.py,sha256=56itZW3S1I7ZYvbxPmFIVPAh1AIJZdljByz1uCrHqFE,635
+sonusai/metrics/__init__.py,sha256=GEenHTytN-_qMAF7UJypqkArudJsHyctGSKd0bKUniQ,807
+sonusai/metrics/calc_audio_stats.py,sha256=u-HIyrqJUCb0G7WR865TcpTVkT2lsCnb2rwB67IgzEc,1308
 sonusai/metrics/calc_class_weights.py,sha256=dyY7daEIf5Ms5tfTf6wF0fkx_GnMADHOZR_rtsfGoVM,3933
 sonusai/metrics/calc_optimal_thresholds.py,sha256=9fRfwl-aKAbzHJyqGHv4o8BpZXG9HHB7zUJObHXfYM4,3522
 sonusai/metrics/calc_pcm.py,sha256=n_QoDwdM09HZx0zMbCkJR07e9fCLo03Ilpi1nqdizRg,1906
-sonusai/metrics/calc_pesq.py,sha256=bO2QQXvQOrULTpUkNs0MYgk23tWCJqQ6wH1314jHdz0,853
+sonusai/metrics/calc_pesq.py,sha256=vO3HnQfyzYLy6VILhP5E-Yi6HkI6T9ob_CEfPo6NG50,971
+sonusai/metrics/calc_phase_distance.py,sha256=Oh7UrH4ekm48pIhOnAOmkSr1OWueq2GUlwJheoAQmnk,1905
 sonusai/metrics/calc_sa_sdr.py,sha256=pup1uzHbSqV3JprYCXmu_hbWILqRME5KAnFJGe4WXCY,2566
 sonusai/metrics/calc_sample_weights.py,sha256=0O2EH1-FKlCa0HFgKPUF1BJiknR1hCH7zLbXnoXH7Ag,945
-sonusai/metrics/calc_wer.py,sha256=RP-K6OcTN-7pxBjy9DQzWLBLzIlLVivsgHedtsr3sfY,2499
-sonusai/metrics/calc_wsdr.py,sha256=ug1P8DGSuEvb0OKb97S8ZkXefz7CDUGiAhVG_rt1Wwo,5413
+sonusai/metrics/calc_snr_f.py,sha256=cLRUYeyRuoDtXWOgrgSzFP3EMOBlRaBcPM1BcxpEU_M,1199
+sonusai/metrics/calc_speech.py,sha256=i-BjYiSY-M3YSbKFUOW4m7m4frkF4ikyJgyizhDjGm0,14448
+sonusai/metrics/calc_wer.py,sha256=7VVW_YsbeysKOZOTtvXVAvWi8lb0PZtu-Xg2HiXomWc,2483
+sonusai/metrics/calc_wsdr.py,sha256=9wiCJsGgRGHzBpHv0zSxorMsN_JZNBPyge280Hd1W0w,2531
 sonusai/metrics/class_summary.py,sha256=4Mb25nuk6eqotnQSFMuOQL3zofGcpNXDfDlPa513ZLQ,3286
 sonusai/metrics/confusion_matrix_summary.py,sha256=3qg6TMKjJeHtNjj2YnNjPFSlMrQXt0Zcu1dLkGB_aPU,4001
 sonusai/metrics/one_hot.py,sha256=QSeH_GdqBpOAKLrNnQ8gjcPC-vSdUqC0yPEQueTA6VI,13548
 sonusai/metrics/snr_summary.py,sha256=P4U5_Xr7v9F8kF-rZBnpsVNt3p42rIVS6zmch8yfVfg,5575
-sonusai/mixture/__init__.py,sha256=rFaHyroCf0Fd-SuxmH4dl8xZVjOe8gFjndouv9RtzIE,5347
-sonusai/mixture/audio.py,sha256=2lqy0DtTMTYhX4aAOIvVtLNn6QB5ivTa7cJIaAlbfAg,2385
+sonusai/mixture/__init__.py,sha256=5M8wEFLkySuaFUKFLhXGPs9SNt2fLwsr6QqEYi7-eDY,5358
+sonusai/mixture/audio.py,sha256=2u7nnKdKwXRA6YWVwCurst5D6CQinKZj5jlm9LhXe9c,2445
 sonusai/mixture/augmentation.py,sha256=Blb90tdTwBOj5w9tRcYyS5H67YJuFiXsGqwZWd7ON4g,10468
 sonusai/mixture/class_count.py,sha256=_wFnVl2yEOnbor7pLg7cYOUeX6nioov-03Cv3SEbh2k,996
-sonusai/mixture/config.py,sha256=d2IzZ1samHWGMpoKzSmUwMyAWWhgmyNoxyO8oiUwbsg,22193
-sonusai/mixture/constants.py,sha256=ZRM7Z8e6EwnL9RwaMVxks-QokN9KMWxnQzAf9VNxa9M,1408
-sonusai/mixture/datatypes.py,sha256=uVJtT2sVGS531pSglhaLLF5hZcI3_0oKQOWmMkrCwqo,8253
+sonusai/mixture/config.py,sha256=Fu6-HkenI_h7BdOTuGX9B-vsbLJD5pvTPd_LAncZgI4,22828
+sonusai/mixture/constants.py,sha256=90qaRIEcmIoS3Od5h_UP0_SkkvG2aE_eYPv6WsIktC0,1427
+sonusai/mixture/datatypes.py,sha256=uYNToIngMB4m0pgmaAFj7H27FvrMCiwaAAXDXkvIB6w,9100
 sonusai/mixture/db_datatypes.py,sha256=GDYbcSrlgUJsesiUUNnR4s5aBkMgviiNSQDaBcgYX7I,1428
 sonusai/mixture/eq_rule_is_valid.py,sha256=MpQwRA5M76wSiQWEI1lW2cLFdPaMttBLcQp3tWD8efM,1243
 sonusai/mixture/feature.py,sha256=Rwuf82IoXzhHPGbKYVGcatImF_ssBf_FfvbqghVPXtg,4116
-sonusai/mixture/generation.py,sha256=ohZnhtHIrdQDql2OF703NnhK07Ys-1qAjiwrIql-oMw,42694
+sonusai/mixture/generation.py,sha256=W3n6ipI-dxg4Wj6YBJn8RTpFqkAyIXzxwObeFbSLq08,42801
 sonusai/mixture/helpers.py,sha256=eC9ZysEa-83VLKen_9PKWzr8w9dkHj4lp6rMB2fNLbg,24759
 sonusai/mixture/log_duration_and_sizes.py,sha256=baTUpqyM15wA125jo9E3posmVJUe3WlpksyO6v9Jul0,1347
-sonusai/mixture/mapped_snr_f.py,sha256=Fdf2uw62FvyKvVy5VywaUtPZGO1zCWQsHlte0bwkKPQ,3121
-sonusai/mixture/mixdb.py,sha256=s40-NToxNXz3UtiVZW9chHIV-tpqT6u-GbFPg8LfQoc,51644
-sonusai/mixture/soundfile_audio.py,sha256=mHa5SIXsu_uE0j3DO52GydRJrvWSzU_nII-7YJfQ6Qo,4154
-sonusai/mixture/sox_audio.py,sha256=HT3kYA9TP5QPCuoOJdUMnGVN-qY6q96DGL8zxuog76o,12277
+sonusai/mixture/mixdb.py,sha256=dtEuF2S4wg8G5pl_lkfly5k0kAg_g9xMukVLdb4G8iY,57458
+sonusai/mixture/soundfile_audio.py,sha256=BwO4lftNvrhoPTJERONcrpxSpM2fjO6kL_e5Ylz742A,4220
+sonusai/mixture/sox_audio.py,sha256=Pu5RLE8lKKlGOzZsbfcVzbDvy7YPGBuGpbD_PxowFqA,12511
 sonusai/mixture/sox_augmentation.py,sha256=kBWPrsFk0EBi71nLcKt5v0GA34bY7g9D9x0cEamNWbU,4564
 sonusai/mixture/spectral_mask.py,sha256=8AkCwhy-PSdP1Uri9miKZP-bXFYnFcH_c9xZCGrHavU,2071
 sonusai/mixture/target_class_balancing.py,sha256=NTNiKZH0_PWLooeow0l41CjJKK8ZTMVbUqz9ZkaNtWk,4900
 sonusai/mixture/targets.py,sha256=wyy5vhLhuN-hqBMBGoziVvEJg3FKFvJFgmEE7_LaV2M,7908
 sonusai/mixture/tokenized_shell_vars.py,sha256=zIAFvwP2WSvkMAGY7f3SJ4KLXI6IBT-U_e9ptnoo5Hc,4803
-sonusai/mixture/torchaudio_audio.py,sha256=KhHeOMsjmbwOaAcoKD61aFvYBYSlA8OysfT5iGn45MA,3010
+sonusai/mixture/torchaudio_audio.py,sha256=V7OUqkiwplQoIIMPBRWBHqsuPHAYYByt5QMWx7cPIJI,3062
 sonusai/mixture/torchaudio_augmentation.py,sha256=1vEDHI0caL1vrgoY2lAWe4CiHE2jKRuKKH7x23GHw0w,4390
 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=okFJeOf43NxfdLqWFCBA2pOGqujRlNDYdAcwwR_m8z8,2875
-sonusai/mixture/truth_functions/energy.py,sha256=ydMtMLjMloG76DB30ZHQ5tkBVh4dkMJ82XEhKBokmIk,4281
+sonusai/mixture/truth_functions/energy.py,sha256=xWxjrHM1xPitRFML-X7ruqXMmXVII5MKDjU38WO6bG0,4863
 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=KAsjugDRooOA5BRcHVAbZRgV7l8S5CFg7CZ0XtKZaQ0,5764
+sonusai/mixture/truth_functions/target.py,sha256=gy2OGAV_uIrZunNBRC6kZx3qg48TSzNI_HaXGzPKg3s,5763
 sonusai/mkwav.py,sha256=zfSyIiQTIK3KV9Ij33jkLhhZIMVYqaROcRQ4S7c4sIo,5364
 sonusai/onnx_predict.py,sha256=jSxhD2oFyGSTHOGCXbW4fRT-k4SqKOboK2JaDO-yWcs,8737
 sonusai/plot.py,sha256=ERkmxMM3qjcCDm4LGDQY4fRAncCYAzP7uW8iZ7_brcg,17105
@@ -79,11 +82,11 @@ sonusai/speech/vctk.py,sha256=EAMEBAzjZUI6dw15n-yI2oCN-H4tzM9t4aUVlOxpAbo,1540
 sonusai/speech/voxceleb.py,sha256=aJGN0yDb2LFLmCKmRzmUEjpZWQ-QGWw6XWOpy9967AI,2686
 sonusai/summarize_metric_spenh.py,sha256=OiZe_bhCq5esXNhsOkHDD7g4ssYrpENDHvDVoPzV9iw,1822
 sonusai/tplot.py,sha256=85T6OPZfxVegHBiSuilFpdgCNMEE0VKAuciNy4rCY5Y,14544
-sonusai/utils/__init__.py,sha256=h7QrOyEBMUMoIBFKZpNwDG8Jg-1uw3bs-qflB3CXxhU,2257
+sonusai/utils/__init__.py,sha256=UEM7SkO_xxu0VW-u_zI3e8Xih2CX4aqYjqr61LO5EQA,2271
 sonusai/utils/asl_p56.py,sha256=-bvQpd-jRQVURbkZJpRoyEAq6gTv9Rc3oFDbh5_lcjY,3861
-sonusai/utils/asr.py,sha256=6y6VYJizHpuQ3MgKbEQ4t2gofO-MW6Ez23oAd6d23IE,2920
+sonusai/utils/asr.py,sha256=V0-cW56KS6aOSakSVwQIbhkqEkKf5l8T11b8ciNxWDw,2132
 sonusai/utils/asr_functions/__init__.py,sha256=JyHK67s97bw7QzrlkboWhws4yNytdPatqzLJxfwx-yw,43
-sonusai/utils/asr_functions/aaware_whisper.py,sha256=Ew3zb8OfbxEW7q1s-KA7D5eph4SjVSUAJgiLK-vVqhI,1985
+sonusai/utils/asr_functions/aaware_whisper.py,sha256=NPqc62wKhshYecu368Yvl2xcImp2a_kV3Q229B2gyOI,1992
 sonusai/utils/audio_devices.py,sha256=LgaXTln1oRArBzaet3rZiIO2plgtaThuGBc3sJ_sLlo,1414
 sonusai/utils/braced_glob.py,sha256=Z_XIpPK17QiP1JbzAnUC5w3oyG8ZovoyM22Wh-Q_vWU,1675
 sonusai/utils/calculate_input_shape.py,sha256=63ILxibYKuTQozY83QN8Y2OOhBEbW_1X47Q0askcHDM,984
@@ -116,7 +119,7 @@ sonusai/utils/stratified_shuffle_split.py,sha256=rJNXvBp-GxoKzH3OpL7k0ANSu5xMP2z
 sonusai/utils/write_audio.py,sha256=ZsPGExwM86QHLLN2LOWekK2uAqf5pV_1oRW811p0QAI,840
 sonusai/utils/yes_or_no.py,sha256=eMLXBVH0cEahiXY4W2KNORmwNQ-ba10eRtldh0y4NYg,263
 sonusai/vars.py,sha256=m2AefF0m5bXWGXpJj8Pi42zWL2ydeEj7bkak3GrtMyM,940
-sonusai-0.18.2.dist-info/METADATA,sha256=PNDYtM4HDRWpp3GBcuos7jymdaKRoXOT7DlLn-fs8XE,2591
-sonusai-0.18.2.dist-info/WHEEL,sha256=sP946D7jFCHeNz5Iq4fL4Lu-PrWrFsgfLXbbkciIZwg,88
-sonusai-0.18.2.dist-info/entry_points.txt,sha256=zMNjEphEPO6B3cD1GNpit7z-yA9tUU5-j3W2v-UWstU,92
-sonusai-0.18.2.dist-info/RECORD,,
+sonusai-0.18.4.dist-info/METADATA,sha256=8Q0ANLLD6sjL_HJgqY3rrAEyj6m2emeOgPeraSjnUZ4,2591
+sonusai-0.18.4.dist-info/WHEEL,sha256=sP946D7jFCHeNz5Iq4fL4Lu-PrWrFsgfLXbbkciIZwg,88
+sonusai-0.18.4.dist-info/entry_points.txt,sha256=zMNjEphEPO6B3cD1GNpit7z-yA9tUU5-j3W2v-UWstU,92
+sonusai-0.18.4.dist-info/RECORD,,

sonusai/mixture/mapped_snr_f.py

@@ -1,100 +0,0 @@
-import numpy as np
-
-
-def calculate_snr_f_statistics(truth_f: np.ndarray) -> tuple[np.ndarray, np.ndarray, np.ndarray, np.ndarray]:
-    """Calculate statistics of snr_f truth data.
-
-    For now, includes mean and standard deviation of the raw values (usually energy)
-    and mean and standard deviation of the dB values (10 * log10).
-    """
-    return (
-        calculate_snr_mean(truth_f),
-        calculate_snr_std(truth_f),
-        calculate_snr_db_mean(truth_f),
-        calculate_snr_db_std(truth_f),
-    )
-
-
-def calculate_snr_mean(truth_f: np.ndarray) -> np.ndarray:
-    """Calculate mean of snr_f truth data."""
-    snr_mean = np.zeros(truth_f.shape[1], dtype=np.float32)
-
-    for c in range(truth_f.shape[1]):
-        tmp_truth = truth_f[:, c]
-        tmp = tmp_truth[np.isfinite(tmp_truth)].astype(np.double)
-
-        if len(tmp) == 0:
-            snr_mean[c] = -np.inf
-        else:
-            snr_mean[c] = np.mean(tmp)
-
-    return snr_mean
-
-
-def calculate_snr_std(truth_f: np.ndarray) -> np.ndarray:
-    """Calculate standard deviation of snr_f truth data."""
-    snr_std = np.zeros(truth_f.shape[1], dtype=np.float32)
-
-    for c in range(truth_f.shape[1]):
-        tmp_truth = truth_f[:, c]
-        tmp = tmp_truth[np.isfinite(tmp_truth)].astype(np.double)
-
-        if len(tmp) == 0:
-            snr_std[c] = -np.inf
-        else:
-            snr_std[c] = np.std(tmp, ddof=1)
-
-    return snr_std
-
-
-def calculate_snr_db_mean(truth_f: np.ndarray) -> np.ndarray:
-    """Calculate dB mean of snr_f truth data."""
-    snr_db_mean = np.zeros(truth_f.shape[1], dtype=np.float32)
-
-    for c in range(truth_f.shape[1]):
-        tmp_truth = truth_f[:, c]
-        tmp = tmp_truth[np.isfinite(tmp_truth)].astype(np.double)
-
-        tmp2 = 10 * np.ma.log10(tmp).filled(-np.inf)
-        tmp2 = tmp2[np.isfinite(tmp2)]
-
-        if len(tmp2) == 0:
-            snr_db_mean[c] = -np.inf
-        else:
-            snr_db_mean[c] = np.mean(tmp2)
-
-    return snr_db_mean
-
-
-def calculate_snr_db_std(truth_f: np.ndarray) -> np.ndarray:
-    """Calculate dB standard deviation of snr_f truth data."""
-    snr_db_std = np.zeros(truth_f.shape[1], dtype=np.float32)
-
-    for c in range(truth_f.shape[1]):
-        tmp_truth = truth_f[:, c]
-        tmp = tmp_truth[np.isfinite(tmp_truth)].astype(np.double)
-
-        tmp2 = 10 * np.ma.log10(tmp).filled(-np.inf)
-        tmp2 = tmp2[np.isfinite(tmp2)]
-
-        if len(tmp2) == 0:
-            snr_db_std[c] = -np.inf
-        else:
-            snr_db_std[c] = np.std(tmp2, ddof=1)
-
-    return snr_db_std
-
-
-def calculate_mapped_snr_f(truth_f: np.ndarray, snr_db_mean: np.ndarray, snr_db_std: np.ndarray) -> np.ndarray:
-    """Calculate mapped SNR from standard SNR energy per bin/class."""
-    import scipy.special as sc
-
-    old_err = np.seterr(divide='ignore', invalid='ignore')
-    num = 10 * np.log10(np.double(truth_f)) - np.double(snr_db_mean)
-    den = np.double(snr_db_std) * np.sqrt(2)
-    q = num / den
-    q = np.nan_to_num(q, nan=-np.inf, posinf=np.inf, neginf=-np.inf)
-    mapped_snr_f = 0.5 * (1 + sc.erf(q))
-    np.seterr(**old_err)
-
-    return mapped_snr_f.astype(np.float32)