sonusai 0.20.3__py3-none-any.whl → 1.0.2__py3-none-any.whl

sonusai/genmixdb.py

@@ -5,7 +5,7 @@ usage: genmixdb [-hvmdjn] LOC
 options:
     -h, --help
     -v, --verbose   Be verbose.
-    -m, --mix       ave mixture data. [default: False].
+    -m, --mix       Save mixture data. [default: False].
     -d, --dryrun    Perform a dry run showing the processed config. [default: False].
     -j, --json      Save JSON version of database. [default: False].
     -n, --nopar     Do not run in parallel. [default: False].
@@ -16,120 +16,14 @@ genmixdb creates a database of training and evaluation feature and truth data ge
 choice of audio neural-network feature types that are supported by the Aaware real-time front-end and truth data that is
 synchronized frame-by-frame with the feature data.
 
-Here are some examples:
-
-#### Adding target data
-Suppose you have an audio file which is an example, or target, of what you want to recognize or detect. Of course, for
-training a NN you also need truth data for that file (also called parameters/labels/classes). If you don't already have
-it, genmixdb can create truth using a variety of generation functions on each frame of the feature data. You can also
-select different feature types. Here's an example:
-
-genmixdb target_gfr32ts2
-
-where target_gfr32ts2 contains config.yml with the following inside:
----
-feature: gfr32ts2
-
-targets:
-  - name: data/target.wav
-
-target_augmentations:
-  - normalize: -3.5
-...
-
-The mixture database is written to a SQLite file (mixdb.db) in the same directory that contains the config.yml file.
-
-#### Target data mix with noise and augmentation
-
-genmixdb mix_gfr32ts2.yml
-
-where mix_gfr32ts2.yml contains:
----
-feature: gfr32ts2
-
-targets:
-  - name: data/target.wav
-
-target_augmentations:
-  - normalize: -3.5
-    pitch: [-3, 0, 3]
-    tempo: [0.8, 1, 1.2]
-
-noises:
-  - name: data/noise.wav
-
-noise_augmentations:
-  - normalize: -3.5
-
-snrs:
-  - 20
-...
-
-In this example a time-domain mixture is created and feature data is calculated as specified by 'feature: gfr32ts2'.
-Various feature types are available which vary in spectral and temporal resolution (4 ms or higher), and other feature
-algorithm parameters. The total feature size, dimension, and #frames for mixture is reported in the log file (the log
-file name is genmixdb.log).
-
-Truth (parameters/labels/classes) can be automatically created per feature output frame based on a variety of truth
-generation functions. By default, these are included with the feature data in a single HDF5 output file. By default,
-truth generation is turned on with default settings (see truth section) and a single class, i.e., detecting a single
-type of sound. The truth format is a single float per class representing the probability of activity/presence, and
-multi-class truth is possible by specifying the number of classes and either a scalar index or a vector of indices in
-which to put the truth result. For example, 'num_class: 3' and 'class_indices: [ 2 ]' adds a 1x3 vector to the feature
-data with truth put in index 2 (others would be 0) for data/target.wav being an audio clip from sound type of class 2.
-
-The mixture is created with potential data augmentation functions in the following way:
-1. apply noise augmentation rule
-2. apply target augmentation rule to each target in the mixture (multiple targets may be used in mixup)
-3. adjust noise and target gains for specified SNR
-4. add augmented noise to augmented target(s)
-
-Note: If an impulse response is part of the target augmentation, truth generation is performed on the targets before
-applying the IRs. In this way, the truth is not impacted by the IR.
-
-The mixture length is the length of the longest target in the mixture, and the noise signal is repeated if it is
-shorter, or trimmed if longer.
-
-#### Target and noise using path lists
-
-Target and noise audio is specified as a list containing text files, audio files, and file globs. Text files are
-processed with items on each line where each item can be a text file, an audio file, or a file glob. Each item will be
-searched for audio files which can be WAV, MP3, FLAC, AIFF, or OGG format with any sample rate, bit depth, or channel
-count. All audio files will be converted to 16 kHz, float32, single channel (only the first channel is used) format
-before processing.
-
-For example,
-
-genmixdb dog-bark.yml
-
-where dog-bark.yml contains:
----
-targets:
-  - name: slib/dog-outside/*.wav
-  - name: slib/dog-inside/*.wav
-
-will find all .wav files in the specified directories and process them as targets.
+For details, see sonusai doc.
 
 """
 
-import signal
-
-from sonusai.mixture import Mixture
+from sonusai.datatypes import Mixture
 from sonusai.mixture import MixtureDatabase
 
 
-def signal_handler(_sig, _frame):
-    import sys
-
-    from sonusai import logger
-
-    logger.info("Canceled due to keyboard interrupt")
-    sys.exit(1)
-
-
-signal.signal(signal.SIGINT, signal_handler)
-
-
 def genmixdb(
     location: str,
     save_mix: bool = False,
@@ -142,23 +36,17 @@ def genmixdb(
     from functools import partial
     from random import seed
 
+    import pandas as pd
     import yaml
 
     from sonusai import logger
-    from sonusai.mixture import SAMPLE_BYTES
-    from sonusai.mixture import SAMPLE_RATE
-    from sonusai.mixture import AugmentationRule
+    from sonusai.constants import SAMPLE_BYTES
+    from sonusai.constants import SAMPLE_RATE
     from sonusai.mixture import MixtureDatabase
-    from sonusai.mixture import balance_targets
     from sonusai.mixture import generate_mixtures
-    from sonusai.mixture import get_all_snrs_from_config
-    from sonusai.mixture import get_augmentation_rules
-    from sonusai.mixture import get_augmented_targets
-    from sonusai.mixture import get_impulse_response_files
-    from sonusai.mixture import get_mixups
-    from sonusai.mixture import get_noise_files
-    from sonusai.mixture import get_target_augmentations_for_mixup
-    from sonusai.mixture import get_target_files
+    from sonusai.mixture import get_effect_rules
+    from sonusai.mixture import get_ir_files
+    from sonusai.mixture import get_source_files
     from sonusai.mixture import initialize_db
     from sonusai.mixture import load_config
     from sonusai.mixture import log_duration_and_sizes
@@ -166,22 +54,20 @@ def genmixdb(
     from sonusai.mixture import populate_class_weights_threshold_table
     from sonusai.mixture import populate_impulse_response_file_table
     from sonusai.mixture import populate_mixture_table
-    from sonusai.mixture import populate_noise_file_table
+    from sonusai.mixture import populate_source_file_table
     from sonusai.mixture import populate_spectral_mask_table
-    from sonusai.mixture import populate_target_file_table
     from sonusai.mixture import populate_top_table
     from sonusai.mixture import populate_truth_parameters_table
     from sonusai.mixture import update_mixid_width
-    from sonusai.utils import dataclass_from_dict
     from sonusai.utils import human_readable_size
     from sonusai.utils import par_track
     from sonusai.utils import seconds_to_hms
     from sonusai.utils import track
 
     config = load_config(location)
-    initialize_db(location=location, test=test)
+    initialize_db(location, test)
 
-    mixdb = MixtureDatabase(location=location, test=test)
+    mixdb = MixtureDatabase(location, test)
 
     populate_top_table(location, config, test)
     populate_class_label_table(location, config, test)
@@ -197,148 +83,94 @@ def genmixdb(
         logger.debug(yaml.dump(config))
 
     if logging:
-        logger.info("Collecting targets")
+        logger.info("Collecting sources")
 
-    target_files = get_target_files(config, show_progress=show_progress)
-
-    if len(target_files) == 0:
-        raise RuntimeError("Canceled due to no targets")
+    source_files = get_source_files(config, show_progress)
+    logger.info("")
 
-    populate_target_file_table(location, target_files, test)
+    if len([file for file in source_files if file.category == "primary"]) == 0:
+        raise RuntimeError("Canceled due to no primary sources")
 
-    if logging:
-        logger.debug("List of targets:")
-        logger.debug(yaml.dump([target.name for target in mixdb.target_files], default_flow_style=False))
-        logger.debug("")
+    populate_source_file_table(location, source_files, test)
 
     if logging:
-        logger.info("Collecting noises")
-
-    noise_files = get_noise_files(config, show_progress=show_progress)
-
-    populate_noise_file_table(location, noise_files, test)
+        logger.info("Sources summary")
+        data = {
+            "category": [],
+            "files": [],
+            "size": [],
+            "duration": [],
+        }
+        for category, source_files in mixdb.source_files.items():
+            audio_samples = sum([source.samples for source in source_files])
+            audio_duration = audio_samples / SAMPLE_RATE
+            data["category"].append(category)
+            data["files"].append(mixdb.num_source_files(category))
+            data["size"].append(human_readable_size(audio_samples * SAMPLE_BYTES, 1))
+            data["duration"].append(seconds_to_hms(seconds=audio_duration))
+
+        df = pd.DataFrame(data)
+        logger.info(df.to_string(index=False, header=False))
+        logger.info("")
 
-    if logging:
-        logger.debug("List of noises:")
-        logger.debug(yaml.dump([noise.name for noise in mixdb.noise_files], default_flow_style=False))
-        logger.debug("")
+        for category, files in mixdb.source_files.items():
+            logger.debug(f"List of {category} sources:")
+            logger.debug(yaml.dump([file.name for file in files], default_flow_style=False))
 
     if logging:
         logger.info("Collecting impulse responses")
 
-    impulse_response_files = get_impulse_response_files(config)
+    ir_files = get_ir_files(config, show_progress=show_progress)
+    logger.info("")
 
-    populate_impulse_response_file_table(location, impulse_response_files, test)
+    populate_impulse_response_file_table(location, ir_files, test)
 
     if logging:
         logger.debug("List of impulse responses:")
-        logger.debug(
-            yaml.dump(
-                [entry.file for entry in mixdb.impulse_response_files],
-                default_flow_style=False,
-            )
-        )
+        for idx, file in enumerate(ir_files):
+            logger.debug(f"id: {idx}, name:{file.name}, delay: {file.delay}, tags: [{', '.join(file.tags)}]")
         logger.debug("")
 
     if logging:
-        logger.info("Collecting target augmentations")
-
-    target_augmentations = get_augmentation_rules(
-        rules=config["target_augmentations"], num_ir=mixdb.num_impulse_response_files
-    )
-    mixups = get_mixups(target_augmentations)
+        logger.info("Collecting effects")
 
-    if logging:
-        for mixup in mixups:
-            logger.debug(f"Expanded list of target augmentation rules for mixup of {mixup}:")
-            for target_augmentation in get_target_augmentations_for_mixup(target_augmentations, mixup):
-                ta_dict = target_augmentation.to_dict()
-                del ta_dict["mixup"]
-                logger.debug(f"- {ta_dict}")
-            logger.debug("")
+    rules = get_effect_rules(location, config, test)
 
     if logging:
-        logger.info("Collecting noise augmentations")
-
-    noise_augmentations = get_augmentation_rules(
-        rules=config["noise_augmentations"], num_ir=mixdb.num_impulse_response_files
-    )
+        logger.info("")
+        for category, effect in rules.items():
+            logger.debug(f"List of {category} rules:")
+            logger.debug(yaml.dump([entry.to_dict() for entry in effect], default_flow_style=False))
 
     if logging:
-        logger.debug("Expanded list of noise augmentations:")
-        for noise_augmentation in noise_augmentations:
-            na_dict = noise_augmentation.to_dict()
-            del na_dict["mixup"]
-            logger.debug(f"- {na_dict}")
+        logger.debug("SNRS:")
+        for category, source in config["sources"].items():
+            if category != "primary":
+                logger.debug(f"  {category}")
+                for snr in source["snrs"]:
+                    logger.debug(f"  - {snr}")
+        logger.debug("")
+        logger.debug("Mix Rules:")
+        for category, source in config["sources"].items():
+            if category != "primary":
+                logger.debug(f"  {category}")
+                for mix_rule in source["mix_rules"]:
+                    logger.debug(f"  - {mix_rule}")
         logger.debug("")
-
-    if logging:
-        logger.debug(f"SNRs: {config['snrs']}\n")
-        logger.debug(f"Random SNRs: {config['random_snrs']}\n")
-        logger.debug(f"Noise mix mode: {mixdb.noise_mix_mode}\n")
         logger.debug("Spectral masks:")
         for spectral_mask in mixdb.spectral_masks:
             logger.debug(f"- {spectral_mask}")
         logger.debug("")
 
-    if logging:
-        logger.info("Collecting augmented targets")
-
-    augmented_targets = get_augmented_targets(target_files, target_augmentations, mixups)
-
-    if config["class_balancing"]:
-        class_balancing_augmentation = dataclass_from_dict(AugmentationRule, config["class_balancing_augmentation"])
-        augmented_targets, target_augmentations = balance_targets(
-            augmented_targets=augmented_targets,
-            targets=target_files,
-            target_augmentations=target_augmentations,
-            class_balancing_augmentation=class_balancing_augmentation,  # pyright: ignore [reportArgumentType]
-            num_classes=mixdb.num_classes,
-            num_ir=mixdb.num_impulse_response_files,
-            mixups=mixups,
-        )
-
-    target_audio_samples = sum([targets.samples for targets in mixdb.target_files])
-    target_audio_duration = target_audio_samples / SAMPLE_RATE
-    noise_audio_duration = sum([noises.duration for noises in mixdb.noise_files])
-    noise_audio_samples = noise_audio_duration * SAMPLE_RATE
-
-    if logging:
-        logger.info("")
-        logger.info(
-            f"Target audio: {mixdb.num_target_files} files, "
-            f"{human_readable_size(target_audio_samples * SAMPLE_BYTES, 1)}, "
-            f"{seconds_to_hms(seconds=target_audio_duration)}"
-        )
-        logger.info(
-            f"Noise audio: {mixdb.num_noise_files} files, "
-            f"{human_readable_size(noise_audio_samples * SAMPLE_BYTES, 1)}, "
-            f"{seconds_to_hms(seconds=noise_audio_duration)}"
-        )
-
     if logging:
         logger.info("Generating mixtures")
 
-    used_noise_files, used_noise_samples, mixtures = generate_mixtures(
-        noise_mix_mode=mixdb.noise_mix_mode,
-        augmented_targets=augmented_targets,
-        target_files=target_files,
-        target_augmentations=target_augmentations,
-        noise_files=noise_files,
-        noise_augmentations=noise_augmentations,
-        spectral_masks=mixdb.spectral_masks,
-        all_snrs=get_all_snrs_from_config(config),
-        mixups=mixups,
-        num_classes=mixdb.num_classes,
-        feature_step_samples=mixdb.feature_step_samples,
-        num_ir=mixdb.num_impulse_response_files,
-    )
+    mixtures = generate_mixtures(location, config, rules, test)
 
     num_mixtures = len(mixtures)
     update_mixid_width(location, num_mixtures, test)
 
     if logging:
-        logger.info("")
         logger.info(f"Found {num_mixtures:,} mixtures to process")
 
     total_duration = float(sum([mixture.samples for mixture in mixtures])) / SAMPLE_RATE
@@ -346,7 +178,6 @@ def genmixdb(
     if logging:
         log_duration_and_sizes(
             total_duration=total_duration,
-            num_classes=mixdb.num_classes,
             feature_step_samples=mixdb.feature_step_samples,
             feature_parameters=mixdb.feature_parameters,
             stride=mixdb.fg_stride,
@@ -386,27 +217,17 @@ def genmixdb(
         show_progress=show_progress,
     )
 
-    total_noise_files = len(noise_files)
-
-    total_samples = mixdb.total_samples()
-    total_duration = float(total_samples / SAMPLE_RATE)
-
-    noise_files_percent = (float(used_noise_files) / float(total_noise_files)) * 100
-    noise_samples_percent = (float(used_noise_samples) / float(noise_audio_samples)) * 100
+    total_duration = float(mixdb.total_samples() / SAMPLE_RATE)
 
     if logging:
         log_duration_and_sizes(
             total_duration=total_duration,
-            num_classes=mixdb.num_classes,
             feature_step_samples=mixdb.feature_step_samples,
             feature_parameters=mixdb.feature_parameters,
             stride=mixdb.fg_stride,
             desc="Actual",
         )
         logger.info("")
-        logger.info(f"Used {noise_files_percent:,.0f}% of noise files")
-        logger.info(f"Used {noise_samples_percent:,.0f}% of noise audio")
-        logger.info("")
 
     if not test and save_json:
         if logging:
@@ -434,12 +255,12 @@ def _process_mixture(
 
     if save_mix:
         write(
-            items=[
-                ("targets", genmix_data.targets),
-                ("target", genmix_data.target),
-                ("noise", genmix_data.noise),
-                ("mixture", genmix_data.mixture),
-            ]
+            items={
+                "sources": genmix_data.sources,
+                "source": genmix_data.source,
+                "noise": genmix_data.noise,
+                "mixture": genmix_data.mixture,
+            }
         )
 
         write_mixture_metadata(mixdb, mixture=mixture)
@@ -450,10 +271,10 @@ def _process_mixture(
 def main() -> None:
     from docopt import docopt
 
-    import sonusai
+    from sonusai import __version__ as sai_version
     from sonusai.utils import trim_docstring
 
-    args = docopt(trim_docstring(__doc__), version=sonusai.__version__, options_first=True)
+    args = docopt(trim_docstring(__doc__), version=sai_version, options_first=True)
 
     import time
     from os import makedirs
@@ -498,17 +319,13 @@ def main() -> None:
     logger.info(f"Creating mixture database for {location}")
     logger.info("")
 
-    try:
-        genmixdb(
-            location=location,
-            save_mix=save_mix,
-            show_progress=True,
-            save_json=save_json,
-            no_par=no_par,
-        )
-    except Exception as e:
-        logger.debug(e)
-        raise
+    genmixdb(
+        location=location,
+        save_mix=save_mix,
+        show_progress=True,
+        save_json=save_json,
+        no_par=no_par,
+    )
 
     end_time = time.monotonic()
     logger.info(f"Completed in {seconds_to_hms(seconds=end_time - start_time)}")
@@ -516,4 +333,11 @@ def main() -> None:
 
 
 if __name__ == "__main__":
-    main()
+    from sonusai import exception_handler
+    from sonusai.utils import register_keyboard_interrupt
+
+    register_keyboard_interrupt()
+    try:
+        main()
+    except Exception as e:
+        exception_handler(e)

sonusai/ir_metric.py

@@ -1,10 +1,9 @@
 """sonusai ir_metric
 
-usage: ir_metric [-hv] [-n NCPU] IRLOC
+usage: ir_metric [-h] [-n NCPU] IRLOC
 
 options:
     -h, --help
-    -v, --verbose               Be verbose.
     -n, --num_process NCPU      Number of parallel processes to use [default: auto]
 
 Calculate delay and gain metrics of impulse response (IR) files <filename>.wav in IRLOC.
@@ -22,7 +21,6 @@ IRLOC  directory containing impulse response data in audio files (.wav, .flac, e
 """
 
 import glob
-import signal
 from os.path import abspath
 from os.path import basename
 from os.path import commonprefix
@@ -42,18 +40,6 @@ from numpy import fft
 from sonusai.utils import braced_iglob
 
 
-def signal_handler(_sig, _frame):
-    import sys
-
-    from sonusai import logger
-
-    logger.info("Canceled due to keyboard interrupt")
-    sys.exit(1)
-
-
-signal.signal(signal.SIGINT, signal_handler)
-
-
 def tdoa(signal, reference, interp=1, phat=False, fs=1, t_max=None):
     """
     Estimates the shift of array signal with respect to reference
@@ -263,7 +249,7 @@ def measure_rt60(h, fs=1, decay_db=60, energy_thres=1.0, plot=False, rt60_tgt=No
         t60_decay = 3 * i_decay20db / fs
         rt60 = t60_decay - t_5db
 
-    # # extropolate to compute the rt60 decay time from decay_db decay time
+    # # extrapolate to compute the rt60 decay time from decay_db decay time
     # decay_time = t_decay - t_5db
     # est_rt60 = (60 / decay_db) * decay_time
 
@@ -297,27 +283,30 @@ def measure_rt60(h, fs=1, decay_db=60, energy_thres=1.0, plot=False, rt60_tgt=No
     return rt60, edt, rt10, rt20, floor
 
 
-def process_path(path, extlist=[".wav", ".WAV", ".flac", ".FLAC", ".mp3", ".aac"]):
+def process_path(path: str, extensions: list[str] | None = None) -> tuple[list, str | None]:
     """
     Check path which can be a single file, a subdirectory, or a regex
     return:
       - a list of files with matching extensions to any in extlist provided (i.e. ['.wav', '.mp3', '.acc'])
       - the basedir of the path, if
     """
+    if extensions is None:
+        extensions = [".wav", ".WAV", ".flac", ".FLAC", ".mp3", ".aac"]
+
     # Check if the path is a single file, and return it as a list with the dirname
     if isfile(path):
-        if any(path.endswith(ext) for ext in extlist):
+        if any(path.endswith(ext) for ext in extensions):
             basedir = dirname(path)  # base directory
             if not basedir:
                 basedir = "./"
             return [path], basedir
-        else:
-            return [], []
+
+        return [], None
 
     # Check if the path is a dir, recursively find all files any of the specified extensions, return file list and dir
     if isdir(path):
         matching_files = []
-        for ext in extlist:
+        for ext in extensions:
             matching_files.extend(glob.glob(join(path, "**/*" + ext), recursive=True))
         return matching_files, path
 
@@ -326,11 +315,12 @@ def process_path(path, extlist=[".wav", ".WAV", ".flac", ".FLAC", ".mp3", ".aac"
     matching_files = []
     for file in braced_iglob(pathname=apath, recursive=True):
         matching_files.append(file)
+
     if matching_files:
         basedir = commonprefix(matching_files)  # Find basedir
         return matching_files, basedir
-    else:
-        return [], []
+
+    return [], None
 
 
 def _process_ir(pfile: str, irtab_col: list, basedir: str) -> pd.DataFrame:
@@ -424,20 +414,19 @@ def _process_ir(pfile: str, irtab_col: list, basedir: str) -> pd.DataFrame:
 def main():
     from docopt import docopt
 
-    import sonusai
-    from sonusai.utils import trim_docstring
+    from . import __version__ as sai_version
+    from .utils.docstring import trim_docstring
 
-    args = docopt(trim_docstring(__doc__), version=sonusai.__version__, options_first=True)
+    args = docopt(trim_docstring(__doc__), version=sai_version, options_first=True)
 
-    verbose = args["--verbose"]
     ir_location = args["IRLOC"]
     num_proc = args["--num_process"]
 
     import psutil
 
-    from sonusai.utils import create_timestamp
-    from sonusai.utils import par_track
-    from sonusai.utils import track
+    from .utils.create_timestamp import create_timestamp
+    from .utils.parallel import par_track
+    from .utils.parallel import track
 
     # Check location, default ext are ['.wav', '.WAV', '.flac', '.FLAC', '.mp3', '.aac']
     pfiles, basedir = process_path(ir_location)
@@ -446,15 +435,15 @@ def main():
     if pfiles is None or len(pfiles) < 1:
         print(f"No IR audio files found in {ir_location}, exiting ...")
         raise SystemExit(1)
-    elif len(pfiles) == 1:
+    if len(pfiles) == 1:
         print(f"Found single IR audio file {ir_location} , writing to *-irmetric.txt ...")
         fbase, ext = splitext(basename(pfiles[0]))
         wlcsv_name = None
         txt_fname = str(join(basedir, fbase + "-irmetric.txt"))
-    elif len(pfiles) > 1:
+    else:
         print(f"Found {len(pfiles)} files under {basedir} for impulse response metric calculations")
-        txt_fname = str(join(basedir, "ir_metric_summary.txt"))
         wlcsv_name = str(join(basedir, "ir_metric_list.csv"))
+        txt_fname = str(join(basedir, "ir_metric_summary.txt"))
 
     num_cpu = psutil.cpu_count()
     cpu_percent = psutil.cpu_percent(interval=1)
@@ -552,4 +541,11 @@ def main():
 
 
 if __name__ == "__main__":
-    main()
+    from sonusai import exception_handler
+    from sonusai.utils import register_keyboard_interrupt
+
+    register_keyboard_interrupt()
+    try:
+        main()
+    except Exception as e:
+        exception_handler(e)