sonusai 0.17.2__tar.gz → 0.18.0__tar.gz

{sonusai-0.17.2 → sonusai-0.18.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: sonusai
-Version: 0.17.2
+Version: 0.18.0
 Summary: Framework for building deep neural network models for sound, speech, and voice AI
 Home-page: https://aaware.com
 License: GPL-3.0-only
@@ -21,12 +21,15 @@ Requires-Dist: h5py (>=3.11.0,<4.0.0)
 Requires-Dist: jiwer (>=3.0.3,<4.0.0)
 Requires-Dist: librosa (>=0.10.1,<0.11.0)
 Requires-Dist: matplotlib (>=3.8.0,<4.0.0)
+Requires-Dist: mgzip (>=0.2.1,<0.3.0)
+Requires-Dist: numpy (>=1.26.4,<2.0.0)
 Requires-Dist: onnx (>=1.14.1,<2.0.0)
 Requires-Dist: onnxruntime (>=1.16.1,<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: praatio (>=6.2.0,<7.0.0)
+Requires-Dist: psutil (>=5,<6)
 Requires-Dist: pyaaware (>=1.5.7,<2.0.0)
 Requires-Dist: pyaudio (>=0.2.14,<0.3.0)
 Requires-Dist: pydub (>=0.25.1,<0.26.0)

{sonusai-0.17.2 → sonusai-0.18.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "sonusai"
-version = "0.17.2"
+version = "0.18.0"
 description = "Framework for building deep neural network models for sound, speech, and voice AI"
 authors = ["Chris Eddington <chris@aaware.com>", "Jason Calderwood <jason@aaware.com>"]
 maintainers = ["Chris Eddington <chris@aaware.com>", "Jason Calderwood <jason@aaware.com>"]
@@ -21,12 +21,15 @@ h5py = "^3.11.0"
 jiwer = "^3.0.3"
 librosa = "^0.10.1"
 matplotlib = "^3.8.0"
+mgzip = "^0.2.1"
+numpy = "^1.26.4"
 onnx = "^1.14.1"
 onnxruntime = "^1.16.1"
 paho-mqtt = "^2.0.0"
 pandas = "^2.1.1"
 pesq = "^0.0.4"
 praatio = "^6.2.0"
+psutil = "^5"
 pyaaware = "^1.5.7"
 pyaudio = "^0.2.14"
 pydub = "^0.25.1"
@@ -47,7 +50,8 @@ mypy = "^1.6.0"
 mypy-extensions = "^1.0.0"
 pytest = "^8.1.1"
 sonusai-asr-cloud = "^0.1.0"
-sonusai-torchl = "^0.1.0"
+sonusai-asr-sensory = "^0.1.0"
+sonusai-torchl = "^0.3.0"
 types-pyyaml = "^6.0.12.12"
 types-requests = "^2.31.0.8"
 

{sonusai-0.17.2 → sonusai-0.18.0}/sonusai/__init__.py

@@ -14,7 +14,6 @@ commands_doc = """
    genmixdb                     Generate a mixture database
    gentcst                      Generate target configuration from a subdirectory tree
    lsdb                         List information about a mixture database
-   mkmanifest                   Make ASR manifest JSON file
    mkwav                        Make WAV files from a mixture database
    onnx_predict                 Run ONNX predict on a trained model
    plot                         Plot mixture data

{sonusai-0.17.2 → sonusai-0.18.0}/sonusai/audiofe.py

@@ -86,7 +86,7 @@ def main() -> None:
     from sonusai.utils import create_timestamp
     from sonusai.utils import get_input_devices
     from sonusai.utils import load_ort_session
-    from sonusai.utils import write_wav
+    from sonusai.utils import write_audio
 
     ts = create_timestamp()
     capture_name = f'audiofe_capture_{ts}'
@@ -121,7 +121,7 @@ def main() -> None:
             logger.exception(e)
             return
         # Only write if capture from device, not for file input
-        write_wav(capture_wav, capture_audio, SAMPLE_RATE)
+        write_audio(capture_wav, capture_audio, SAMPLE_RATE)
         logger.info('')
         logger.info(f'Wrote capture audio with shape {capture_audio.shape} to {capture_wav}')
 
@@ -175,7 +175,7 @@ def main() -> None:
             logger.info(f'Wrote predict with shape {predict.shape} to {h5_name}')
 
         predict_audio = get_audio_from_feature(feature=predict, feature_mode=feature_mode)
-        write_wav(predict_wav, predict_audio, SAMPLE_RATE)
+        write_audio(predict_wav, predict_audio, SAMPLE_RATE)
         logger.info(f'Wrote predict audio with shape {predict_audio.shape} to {predict_wav}')
         if debug:
             with h5py.File(h5_name, 'a') as f:

{sonusai-0.17.2 → sonusai-0.18.0}/sonusai/calc_metric_spenh.py

@@ -1,6 +1,6 @@
 """sonusai calc_metric_spenh
 
-usage: calc_metric_spenh [-hvtpws] [-i MIXID] [-e ASR] [-m MODEL] PLOC TLOC
+usage: calc_metric_spenh [-hvtpws] [-i MIXID] [-e ASR] [-m MODEL] [-n NCPU] PLOC TLOC
 
 options:
     -h, --help
@@ -10,6 +10,7 @@ options:
     -p, --plot                  Enable PDF plots file generation per mixture.
     -w, --wav                   Generate WAV files per mixture.
     -s, --summary               Enable summary files generation.
+    -n, --num_process NCPU      Number of parallel processes to use [default: auto]
     -e ASR, --asr-method ASR    ASR method: deepgram, google, aixplain_whisper, whisper, or sensory. [default: none]
     -m MODEL, --model           ASR model name used in some ASR methods. [default: tiny]
 
@@ -154,8 +155,8 @@ def snr(clean_speech, processed_speech, sample_rate):
         signal_energy = np.sum(np.square(clean_frame))
         noise_energy = np.sum(np.square(clean_frame - processed_frame))
         segmental_snr[frame_count] = 10 * np.log10(signal_energy / (noise_energy + eps) + eps)
-        segmental_snr[frame_count] = np.max(segmental_snr[frame_count], min_snr)
-        segmental_snr[frame_count] = np.min(segmental_snr[frame_count], max_snr)
+        segmental_snr[frame_count] = max(segmental_snr[frame_count], min_snr)
+        segmental_snr[frame_count] = min(segmental_snr[frame_count], max_snr)
 
         start = start + skip_rate
 
@@ -697,11 +698,14 @@ def plot_e_predict_truth(predict: np.ndarray,
 
 
 def _process_mixture(mixid: int) -> tuple[pd.DataFrame, pd.DataFrame]:
+    import pickle
     from os.path import basename
     from os.path import join
     from os.path import splitext
 
     import h5py
+    import mgzip
+    from matplotlib.backends.backend_pdf import PdfPages
     from numpy import inf
     from pystoi import stoi
 
@@ -718,7 +722,7 @@ def _process_mixture(mixid: int) -> tuple[pd.DataFrame, pd.DataFrame]:
     from sonusai.utils import reshape_outputs
     from sonusai.utils import stack_complex
     from sonusai.utils import unstack_complex
-    from sonusai.utils import write_wav
+    from sonusai.utils import write_audio
 
     mixdb = MP_GLOBAL.mixdb
     predict_location = MP_GLOBAL.predict_location
@@ -800,8 +804,12 @@ def _process_mixture(mixid: int) -> tuple[pd.DataFrame, pd.DataFrame]:
             mixture = mixture[0:-trim_t]
             truth_f = truth_f[0:-trim_f, :]
         elif predict.shape[0] > target_f.shape[0]:
-            raise SonusAIError(
-                f'Error: prediction has more frames than true mixture {predict.shape[0]} vs {truth_f.shape[0]}')
+            logger.debug(
+                f'Warning: prediction has more frames than true mixture {predict.shape[0]} vs {truth_f.shape[0]}')
+            trim_f = predict.shape[0] - target_f.shape[0]
+            predict = predict[0:-trim_f, :]
+            # raise SonusAIError(
+            #     f'Error: prediction has more frames than true mixture {predict.shape[0]} vs {truth_f.shape[0]}')
 
     # 3) Extraction - format proper complex and wav estimates and truth (unstack, uncompress, inv tf, etc.)
     if truth_est_mode:
@@ -883,13 +891,9 @@ def _process_mixture(mixid: int) -> tuple[pd.DataFrame, pd.DataFrame]:
         wer_tge = float('nan')
         wer_pi = float('nan')
     else:
-        asr_tt = MP_GLOBAL.mixdb.get_speech_metadata(mixid, 'text')[0]  # ignore mixup
+        asr_tt = MP_GLOBAL.mixdb.mixture_speech_metadata(mixid, 'text')[0]  # ignore mixup
         if asr_tt is None:
             asr_tt = calc_asr(target, engine=asr_method, whisper_model_name=asr_model_name).text  # target truth
-        # if MP_GLOBAL.mixdb.asr_manifests:
-        #     asr_tt = MP_GLOBAL.mixdb.mixture_asr_data(mixid)[0]  # ignore mixup
-        # else:
-        #     asr_tt = calc_asr(target, engine=asr_method, whisper_model_name=asr_model_name).text  # target truth
 
         if asr_tt:
             asr_mx = calc_asr(mixture, engine=asr_method, whisper_model_name=asr_model_name).text
@@ -957,10 +961,7 @@ def _process_mixture(mixid: int) -> tuple[pd.DataFrame, pd.DataFrame]:
         print(f'Noise path: {mixdb.noise_file(ni).name}', file=f)
         if asr_method != 'none':
             print(f'ASR method: {asr_method} and whisper model (if used):  {asr_model_name}', file=f)
-            if mixdb.asr_manifests:
-                print(f'ASR truth from metadata:  {asr_tt}', file=f)
-            else:
-                print(f'ASR truth from wer method:  {asr_tt}', file=f)
+            print(f'ASR truth:  {asr_tt}', file=f)
             print(f'ASR result for mixture:  {asr_mx}', file=f)
             print(f'ASR result for prediction:  {asr_tge}', file=f)
 
@@ -968,12 +969,12 @@ def _process_mixture(mixid: int) -> tuple[pd.DataFrame, pd.DataFrame]:
 
     # 7) write wav files
     if enable_wav:
-        write_wav(name=base_name + '_mixture.wav', audio=float_to_int16(mixture))
-        write_wav(name=base_name + '_target.wav', audio=float_to_int16(target))
-        # write_wav(name=base_name + '_target_fi.wav', audio=float_to_int16(target_fi))
-        write_wav(name=base_name + '_noise.wav', audio=float_to_int16(noise))
-        write_wav(name=base_name + '_target_est.wav', audio=float_to_int16(target_est_wav))
-        write_wav(name=base_name + '_noise_est.wav', audio=float_to_int16(noise_est_wav))
+        write_audio(name=base_name + '_mixture.wav', audio=float_to_int16(mixture))
+        write_audio(name=base_name + '_target.wav', audio=float_to_int16(target))
+        # write_audio(name=base_name + '_target_fi.wav', audio=float_to_int16(target_fi))
+        write_audio(name=base_name + '_noise.wav', audio=float_to_int16(noise))
+        write_audio(name=base_name + '_target_est.wav', audio=float_to_int16(target_est_wav))
+        write_audio(name=base_name + '_noise_est.wav', audio=float_to_int16(noise_est_wav))
 
         # debug code to test for perfect reconstruction of the extraction method
         # note both 75% olsa-hanns and 50% olsa-hann modes checked to have perfect reconstruction
@@ -984,7 +985,6 @@ def _process_mixture(mixid: int) -> tuple[pd.DataFrame, pd.DataFrame]:
 
     # 8) Write out plot file
     if enable_plot:
-        from matplotlib.backends.backend_pdf import PdfPages
         plot_name = base_name + '_metric_spenh.pdf'
 
         # Reshape feature to eliminate overlap redundancy for easier to understand spectrogram view
@@ -1015,12 +1015,15 @@ def _process_mixture(mixid: int) -> tuple[pd.DataFrame, pd.DataFrame]:
                 tfunc_name = tfunc_name + ' (db)'
 
             mixspec = 20 * np.log10(abs(mixture_f) + np.finfo(np.float32).eps)
-            pdf.savefig(plot_mixpred(mixture=mixture,
-                                     mixture_f=mixspec,
-                                     target=target,
-                                     feature=feat_sgram,
-                                     predict=predplot,
-                                     tp_title=tfunc_name))
+            fig_obj = plot_mixpred(mixture=mixture,
+                                   mixture_f=mixspec,
+                                   target=target,
+                                   feature=feat_sgram,
+                                   predict=predplot,
+                                   tp_title=tfunc_name)
+            pdf.savefig(fig_obj)
+            with mgzip.open(base_name + '_metric_spenh_fig1.mfigz', 'wb') as f:
+                pickle.dump(fig_obj, f)
 
             # ----- page 2, plot unmapped predict, opt truth reconstructed and line plots of mean-over-f
             # pdf.savefig(plot_pdb_predtruth(predict=pred_snr_f, tp_title='predict snr_f (db)'))
@@ -1029,22 +1032,28 @@ def _process_mixture(mixid: int) -> tuple[pd.DataFrame, pd.DataFrame]:
             tg_spec = 20 * np.log10(abs(target_f) + np.finfo(np.float32).eps)
             tg_est_spec = 20 * np.log10(abs(predict_complex) + np.finfo(np.float32).eps)
             # n_spec = np.reshape(n_spec,(n_spec.shape[0] * n_spec.shape[1], n_spec.shape[2]))
-            pdf.savefig(plot_e_predict_truth(predict=tg_est_spec,
-                                             predict_wav=target_est_wav,
-                                             truth_f=tg_spec,
-                                             truth_wav=target_fi,
-                                             metric=np.vstack((lerr_tg_frame, phd_frame)).T,
-                                             tp_title='speech estimate'))
+            fig_obj = plot_e_predict_truth(predict=tg_est_spec,
+                                           predict_wav=target_est_wav,
+                                           truth_f=tg_spec,
+                                           truth_wav=target_fi,
+                                           metric=np.vstack((lerr_tg_frame, phd_frame)).T,
+                                           tp_title='speech estimate')
+            pdf.savefig(fig_obj)
+            with mgzip.open(base_name + '_metric_spenh_fig2.mfigz', 'wb') as f:
+                pickle.dump(fig_obj, f)
 
             # page 4 noise extraction
             n_spec = 20 * np.log10(abs(noise_f) + np.finfo(np.float32).eps)
             n_est_spec = 20 * np.log10(abs(noise_est_complex) + np.finfo(np.float32).eps)
-            pdf.savefig(plot_e_predict_truth(predict=n_est_spec,
-                                             predict_wav=noise_est_wav,
-                                             truth_f=n_spec,
-                                             truth_wav=noise_fi,
-                                             metric=lerr_n_frame,
-                                             tp_title='noise estimate'))
+            fig_obj = plot_e_predict_truth(predict=n_est_spec,
+                                           predict_wav=noise_est_wav,
+                                           truth_f=n_spec,
+                                           truth_wav=noise_fi,
+                                           metric=lerr_n_frame,
+                                           tp_title='noise estimate')
+            pdf.savefig(fig_obj)
+            with mgzip.open(base_name + '_metric_spenh_fig4.mfigz', 'wb') as f:
+                pickle.dump(fig_obj, f)
 
             # Plot error waveforms
             # tg_err_wav = target_fi - target_est_wav
@@ -1072,6 +1081,7 @@ def main():
     enable_wav = args['--wav']
     enable_summary = args['--summary']
     predict_location = args['PLOC']
+    num_proc = args['--num_process']
     truth_location = args['TLOC']
 
     import glob
@@ -1080,6 +1090,7 @@ def main():
     from os.path import join
     from os.path import split
 
+    import psutil
     from tqdm import tqdm
 
     from sonusai import create_file_handler
@@ -1153,13 +1164,17 @@ def main():
         fnb = 'metric_spenh_fwhsp_' + asr_model_name + '_'
         logger.info(f'ASR enabled with method {asr_method} and whisper model {asr_model_name}')
         enable_asr_warmup = True
+    elif asr_method == 'sensory':
+        fnb = 'metric_spenh_snsr_' + asr_model_name + '_'
+        logger.info(f'ASR enabled with method {asr_method} and model {asr_model_name}')
+        enable_asr_warmup = True
     else:
         logger.error(f'Unrecognized ASR method: {asr_method}')
         return
 
     if enable_asr_warmup:
-        DEFAULT_SPEECH = split(DEFAULT_NOISE)[0] + '/speech_ma01_01.wav'
-        audio = read_audio(DEFAULT_SPEECH)
+        default_speech = split(DEFAULT_NOISE)[0] + '/speech_ma01_01.wav'
+        audio = read_audio(default_speech)
         logger.info(f'Warming up asr method, note for cloud service this could take up to a few min ...')
         asr_chk = calc_asr(audio, engine=asr_method, whisper_model_name=asr_model_name)
         logger.info(f'Warmup completed, results {asr_chk}')
@@ -1173,10 +1188,25 @@ def main():
     MP_GLOBAL.asr_method = asr_method
     MP_GLOBAL.asr_model_name = asr_model_name
 
+    num_cpu = psutil.cpu_count()
+    cpu_percent = psutil.cpu_percent(interval=1)
+    logger.info(f"#CPUs: {num_cpu}, current CPU utilization: {cpu_percent}%")
+    logger.info(f"Memory utilization: {psutil.virtual_memory().percent}%")
+    if num_proc == 'auto':
+        use_cpu = int(num_cpu * (0.9 - cpu_percent / 100))  # default use 80% of available cpus
+    elif num_proc == 'None':
+        use_cpu = None
+    else:
+        use_cpu = min(max(int(num_proc), 1), num_cpu)
+
     # Individual mixtures use pandas print, set precision to 2 decimal places
     # pd.set_option('float_format', '{:.2f}'.format)
+    logger.info(f"Calculating metrics for {len(mixids)} mixtures using {use_cpu} parallel processes ...")
     progress = tqdm(total=len(mixids), desc='calc_metric_spenh')
-    all_metrics_tables = pp_tqdm_imap(_process_mixture, mixids, progress=progress, num_cpus=8)
+    if use_cpu is None:
+        all_metrics_tables = pp_tqdm_imap(_process_mixture, mixids, progress=progress, no_par=True)
+    else:
+        all_metrics_tables = pp_tqdm_imap(_process_mixture, mixids, progress=progress, num_cpus=use_cpu)
     progress.close()
 
     all_metrics_table_1 = pd.concat([item[0] for item in all_metrics_tables])
@@ -1209,9 +1239,9 @@ def main():
     for i in range(len(mtab_snr_summary)):
         if mtab_snr_summary['MXWER'].iloc[i] == 0.0:
             if mtab_snr_summary['WER'].iloc[i] == 0.0:
-                mtab_snr_summary['WERi%'].iloc[i] = 0.0
+                mtab_snr_summary.iloc[i, 6] = 0.0  # mtab_snr_summary['WERi%'].iloc[i] = 0.0
             else:
-                mtab_snr_summary['WERi%'].iloc[i] = -999.0
+                mtab_snr_summary.iloc[i, 6] = -999.0  # mtab_snr_summary['WERi%'].iloc[i] = -999.0
         else:
             if ~np.isnan(mtab_snr_summary['WER'].iloc[i]) and ~np.isnan(mtab_snr_summary['MXWER'].iloc[i]):
                 # update WERi% in 6th col
@@ -1240,7 +1270,6 @@ def main():
     if num_mix > 1:
         # Print pandas data to files using precision to 2 decimals
         # pd.set_option('float_format', '{:.2f}'.format)
-        csp = 0
 
         if not truth_est_mode:
             ofname = join(predict_location, fnb + 'summary.txt')
@@ -1280,9 +1309,9 @@ def main():
 
         # Write summary to .csv file
         if not truth_est_mode:
-            csv_name = join(predict_location, fnb + 'summary.csv')
+            csv_name = str(join(predict_location, fnb + 'summary.csv'))
         else:
-            csv_name = join(predict_location, fnb + 'summary_truest.csv')
+            csv_name = str(join(predict_location, fnb + 'truest_summary.csv'))
         header_args = {
             'mode':     'a',
             'encoding': 'utf-8',
@@ -1315,16 +1344,16 @@ def main():
         pd.DataFrame([label]).to_csv(csv_name, **header_args)
 
         if not truth_est_mode:
-            csv_name = join(predict_location, fnb + 'list.csv')
+            csv_name = str(join(predict_location, fnb + 'list.csv'))
         else:
-            csv_name = join(predict_location, fnb + 'list_truest.csv')
+            csv_name = str(join(predict_location, fnb + 'truest_list.csv'))
         pd.DataFrame(['Speech enhancement metrics list:']).to_csv(csv_name, header=False, index=False)  # open as write
         all_metrics_table_1.round(2).to_csv(csv_name, **table_args)
 
         if not truth_est_mode:
-            csv_name = join(predict_location, fnb + 'estats_list.csv')
+            csv_name = str(join(predict_location, fnb + 'estats_list.csv'))
         else:
-            csv_name = join(predict_location, fnb + 'estats_list_truest.csv')
+            csv_name = str(join(predict_location, fnb + 'truest_estats_list.csv'))
         pd.DataFrame(['Extraction statistics list:']).to_csv(csv_name, header=False, index=False)  # open as write
         all_metrics_table_2.round(2).to_csv(csv_name, **table_args)
 

{sonusai-0.17.2 → sonusai-0.18.0}/sonusai/doc/doc.py

@@ -255,30 +255,6 @@ The 'truth_settings' parameter specifies the following:
 """ + get_truth_functions() + default
 
 
-def doc_asr_manifest() -> str:
-    default = f"\nDefault value: {get_default_config()['asr_manifest']}"
-    return """
-'asr_manifest' is a mixture database configuration parameter that defines an
-optional ASR manifest.
-
-The parameter takes a list of manifest files to be used to populate ASR data
-per target. Each line of the manifest should be in the following format:
-
-{"audio_filepath": "/path/to/audio.wav", "text": "the transcription of the utterance", "duration": 23.147}
-
-The audio_filepath field should provide an absolute path to the audio file corresponding
-to the utterance. The text field should contain the full transcript for the utterance,
-and the duration field should reflect the duration of the utterance in seconds.
-
-Each entry in the manifest (describing one audio file) should be bordered by '{' and '}'
-and must be contained on one line. The fields that describe the file should be separated
-by commas, and have the form "field_name": value, as shown above.
-
-Since the manifest specifies the path for each utterance, the audio files do not have to be
-located in the same directory as the manifest, or even in any specific directory structure.
-""" + default
-
-
 def doc_augmentations() -> str:
     return """
 Augmentation Rules

sonusai-0.18.0/sonusai/genmetrics.py

@@ -0,0 +1,146 @@
+# Generate mixdb metrics based on metrics listed in config.yml
+
+
+class MixtureMetrics:
+    @property
+    def mxsnr(self):
+        ...
+
+    @property
+    def mxssnravg(self):
+        ...
+
+    @property
+    def mxssnrstd(self):
+        ...
+
+    @property
+    def mxssnrdavg(self):
+        ...
+
+    @property
+    def mxssnrdstd(self):
+        ...
+
+    @property
+    def mxpesq(self):
+        ...
+
+    @property
+    def mxwsdr(self):
+        ...
+
+    @property
+    def mxpd(self):
+        ...
+
+    @property
+    def mxstoi(self):
+        ...
+
+    @property
+    def mxcsig(self):
+        ...
+
+    @property
+    def mxcbak(self):
+        ...
+
+    @property
+    def mxcovl(self):
+        ...
+
+    def mxwer(self, engine: str, model: str):
+        ...
+
+    @property
+    def tdco(self):
+        ...
+
+    @property
+    def tmin(self):
+        ...
+
+    @property
+    def tmax(self):
+        ...
+
+    @property
+    def tpkdb(self):
+        ...
+
+    @property
+    def tlrms(self):
+        ...
+
+    @property
+    def tpkr(self):
+        ...
+
+    @property
+    def ttr(self):
+        ...
+
+    @property
+    def tcr(self):
+        ...
+
+    @property
+    def tfl(self):
+        ...
+
+    @property
+    def tpkc(self):
+        ...
+
+    @property
+    def ndco(self):
+        ...
+
+    @property
+    def nmin(self):
+        ...
+
+    @property
+    def nmax(self):
+        ...
+
+    @property
+    def npkdb(self):
+        ...
+
+    @property
+    def nlrms(self):
+        ...
+
+    @property
+    def npkr(self):
+        ...
+
+    @property
+    def ntr(self):
+        ...
+
+    @property
+    def ncr(self):
+        ...
+
+    @property
+    def nfl(self):
+        ...
+
+    @property
+    def npkc(self):
+        ...
+
+    @property
+    def sedavg(self):
+        ...
+
+    @property
+    def sedcnt(self):
+        ...
+
+    @property
+    def sedtopn(self):
+        ...

{sonusai-0.17.2 → sonusai-0.18.0}/sonusai/genmixdb.py

@@ -174,7 +174,6 @@ def genmixdb(location: str,
     from sonusai.mixture import initialize_db
     from sonusai.mixture import load_config
     from sonusai.mixture import log_duration_and_sizes
-    from sonusai.mixture import populate_asr_manifest_table
     from sonusai.mixture import populate_class_label_table
     from sonusai.mixture import populate_class_weights_threshold_table
     from sonusai.mixture import populate_impulse_response_file_table
@@ -195,7 +194,6 @@ def genmixdb(location: str,
     mixdb = MixtureDatabase(location=location, test=test)
 
     populate_top_table(location, config, test)
-    populate_asr_manifest_table(location, config, test)
     populate_class_label_table(location, config, test)
     populate_class_weights_threshold_table(location, config, test)
     populate_spectral_mask_table(location, config, test)

{sonusai-0.17.2 → sonusai-0.18.0}/sonusai/mixture/__init__.py

@@ -88,7 +88,6 @@ from .feature import get_feature_from_audio
 from .generation import generate_mixtures
 from .generation import get_all_snrs_from_config
 from .generation import initialize_db
-from .generation import populate_asr_manifest_table
 from .generation import populate_class_label_table
 from .generation import populate_class_weights_threshold_table
 from .generation import populate_impulse_response_file_table

{sonusai-0.17.2 → sonusai-0.18.0}/sonusai/mixture/constants.py

@@ -4,7 +4,6 @@ from importlib.resources import as_file
 from importlib.resources import files
 
 REQUIRED_CONFIGS = [
-    'asr_manifest',
     'class_balancing',
     'class_balancing_augmentation',
     'class_labels',

{sonusai-0.17.2 → sonusai-0.18.0}/sonusai/mixture/datatypes.py

@@ -1,5 +1,4 @@
 from dataclasses import dataclass
-from dataclasses import field
 from typing import Optional
 from typing import TypeAlias
 
@@ -135,13 +134,7 @@ class UniversalSNR(float):
         return self._is_random
 
 
-# @dataclass(frozen=True)
-# class UniversalSNR:
-#     is_random: bool
-#     value: float
-#
-#     def __lt__(self, other) -> bool:
-#         return self.value < other.value
+Speaker: TypeAlias = dict[str, str]
 
 
 @dataclass
@@ -151,6 +144,7 @@ class TargetFile(DataClassSonusAIMixin):
     truth_settings: TruthSettings
     class_balancing_augmentation: Optional[AugmentationRule] = None
     level_type: Optional[str] = None
+    speaker_id: Optional[int] = None
 
     @property
     def duration(self) -> float:
@@ -317,7 +311,6 @@ class FeatureGeneratorInfo:
 
 @dataclass
 class MixtureDatabaseConfig(DataClassSonusAIMixin):
-    asr_manifest: list[str] = field(default_factory=list)
     class_balancing: Optional[bool] = False
     class_labels: Optional[list[str]] = None
     class_weights_threshold: Optional[list[float]] = None