sonusai 0.12.6__tar.gz → 0.12.7__tar.gz

{sonusai-0.12.6 → sonusai-0.12.7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: sonusai
-Version: 0.12.6
+Version: 0.12.7
 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-0.12.6 → sonusai-0.12.7}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "sonusai"
-version = "0.12.6"
+version = "0.12.7"
 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>"]

{sonusai-0.12.6 → sonusai-0.12.7}/sonusai/calc_metric_spenh.py

@@ -808,7 +808,7 @@ def main() -> None:
     # Individual mixtures use pandas print, set precision to 2 decimal places
     # pd.set_option('float_format', '{:.2f}'.format)
     progress = tqdm(total=len(mixids))
-    all_metrics_tables = p_tqdm_map(_process_mixture, mixids, progress=progress)
+    all_metrics_tables = p_tqdm_map(_process_mixture, mixids, progress=progress, chunksize=10)
     progress.close()
 
     all_metrics_table_1 = pd.concat([item[0] for item in all_metrics_tables])

{sonusai-0.12.6 → sonusai-0.12.7}/sonusai/genft.py

@@ -72,7 +72,7 @@ def genft(mixdb: MixtureDatabase,
             results.append(_genft_kernel(mixid))
     else:
         progress = tqdm(total=len(mixids), disable=not show_progress)
-        results = p_tqdm_map(_genft_kernel, mixids, progress=progress)
+        results = p_tqdm_map(_genft_kernel, mixids, progress=progress, chunksize=10)
         progress.close()
 
     return results

{sonusai-0.12.6 → sonusai-0.12.7}/sonusai/genmix.py

@@ -79,7 +79,7 @@ def genmix(mixdb: MixtureDatabase,
             results.append(_genmix_kernel(mixid))
     else:
         progress = tqdm(total=len(mixids), disable=not show_progress)
-        results = p_tqdm_map(_genmix_kernel, mixids, progress=progress)
+        results = p_tqdm_map(_genmix_kernel, mixids, progress=progress, chunksize=10)
         progress.close()
 
     return results

{sonusai-0.12.6 → sonusai-0.12.7}/sonusai/genmixdb.py

@@ -345,7 +345,7 @@ def genmixdb(location: Location,
     if logging:
         logger.info('Generating mixtures')
     progress = tqdm(total=total_mixtures, disable=not show_progress)
-    mixdb.mixtures = p_tqdm_map(_process_mixture, range(total_mixtures), progress=progress)
+    mixdb.mixtures = p_tqdm_map(_process_mixture, range(total_mixtures), progress=progress, chunksize=10)
     progress.close()
 
     total_samples = mixdb.total_samples()

{sonusai-0.12.6 → sonusai-0.12.7}/sonusai/main.py

@@ -4,7 +4,6 @@ usage: sonusai [--version] [--help] <command> [<args>...]
 
 The sonusai commands are:
    calc_metric_spenh            Run speech enhancement and analysis
-   calc_metric_spenh_targetf    Run speech enhancement and analysis for targetf truth (deprecated)
    evaluate                     Evaluate model performance
    genft                        Generate feature and truth data
    genmix                       Generate mixture and truth data
@@ -14,6 +13,7 @@ The sonusai commands are:
    keras_train                  Train a model using Keras
    keras_onnx                   Convert a trained Keras model to ONNX
    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
@@ -38,7 +38,6 @@ def main() -> None:
 
     commands = (
         'calc_metric_spenh',
-        'calc_metric_spenh_targetf',
         'evaluate',
         'genft',
         'genmix',
@@ -48,6 +47,7 @@ def main() -> None:
         'keras_train',
         'keras_onnx',
         'lsdb',
+        'mkmanifest',
         'mkwav',
         'onnx_predict',
         'plot',

{sonusai-0.12.6 → sonusai-0.12.7}/sonusai/mixture/config.py

@@ -250,7 +250,7 @@ def get_target_files(config: dict, show_progress: bool = False) -> TargetFiles:
                                              for target in config['targets']]))
 
     progress = tqdm(total=len(target_files), disable=not show_progress)
-    target_files = p_tqdm_map(_get_samples, target_files, progress=progress)
+    target_files = p_tqdm_map(_get_samples, target_files, progress=progress, chunksize=10)
     progress.close()
 
     max_class = get_max_class(config['num_classes'], config['truth_mode'] == 'mutex')
@@ -394,7 +394,7 @@ def get_noise_files(config: dict, show_progress: bool = False) -> NoiseFiles:
     noise_files = list(chain.from_iterable([_append_noise_files(noise_file=noise) for noise in config['noises']]))
 
     progress = tqdm(total=len(noise_files), disable=not show_progress)
-    noise_files = p_tqdm_map(_get_samples, noise_files, progress=progress)
+    noise_files = p_tqdm_map(_get_samples, noise_files, progress=progress, chunksize=10)
     progress.close()
 
     return dataclass_from_dict(NoiseFiles, noise_files)

{sonusai-0.12.6 → sonusai-0.12.7}/sonusai/mixture/mixdb.py

@@ -1065,7 +1065,7 @@ class MixtureDatabase:
         :param mixid: Mixture ID
         :param targets: List of augmented target audio data (one per target in the mixup) for the given mixid
         :param noise: Augmented noise audio data for the given mixid
-        :param force: Force computing data from original sources regardless of whether or not cached data exists
+        :param force: Force computing data from original sources regardless of whether cached data exists
         :return: truth_t data
         """
         import numpy as np

sonusai-0.12.7/sonusai/mkmanifest.py

@@ -0,0 +1,174 @@
+"""mkmanifest
+
+usage: mkmanifest [-hvn] [--include GLOB] [-m METHOD] [-e ADAT] [-o OUTPUT] PATH ...
+
+options:
+    -h, --help
+    -v, --verbose                   Be verbose: list all files found.
+    -n, --dry-run                   Collect files, but exit without processing and writing manifest file.
+    --include GLOB                  Search only files whose base name matches GLOB. [default: *.{wav,flac}].
+    -m METHOD, --method METHOD      Method for getting the true speech text of the audio files. [default: librispeech].
+    -e ADAT, --audio-env ADAT       Environment variable pointing to all audio data.
+    -o OUTPUT, --output OUTPUT      Output file name. [default: asr_manifest.json].
+
+Make a speech recognition (ASR) .json manifest file of all audio files under PATHS following the NVIDIA NeMo format.
+An example of manifest entries:
+
+{"audio_filepath": "<absolute_path_to>/1355-39947-0000.wav", "duration": 11.3, "text": "psychotherapy ..."}
+{"audio_filepath": "<absolute_path_to>/1355-39947-0001.wav", "duration": 15.905, "text": "it is an ..."}
+
+See the NVIDIA NeMo docs for more information:
+https://docs.nvidia.com/deeplearning/nemo/user-guide/docs/en/stable/asr/datasets.html
+
+Inputs:
+    PATH        A relative path name or list of paths containing audio files. Each will be
+                recursively searched for files matching the pattern GLOB.
+    GLOB        Match the pattern GLOB using wildcard matching.
+                Example: '*.{wav,flac}' matches all .wav and .flac files.
+    METHOD      The method to use for fetching the true speech of the audio files.
+                Supported methods:
+                    - 'librispeech'
+                    - 'vctk_noisy_speech'
+    ADAT        Audio data environment variable. All found files will be expanded to their full, absolute path and
+                then parts of the path that match the specified environment variable value will be replaced with
+                the variable. This accommodates portability across platforms where the sound datasets may in
+                different locations.
+    OUTPUT      Name of output file. Default is asr_manifest.json.
+
+Outputs the following to the current directory:
+    <OUTPUT>
+    mkmanifest.log
+
+Example usage for LibriSpeech:
+  sonusai mkmanifest -mlibrispeech -eADAT -oasr_manifest.json --include='*.flac' train-clean-100
+
+"""
+from sonusai import logger
+
+VALID_METHOD = ['librispeech', 'vctk_noisy_speech']
+
+
+def main() -> None:
+    from docopt import docopt
+
+    import sonusai
+    from sonusai.utils import trim_docstring
+
+    args = docopt(trim_docstring(__doc__), version=sonusai.__version__, options_first=True)
+
+    verbose = args['--verbose']
+    dry_run = args['--dry-run']
+    include = args['--include']
+    method = args['--method']
+    audio_env = args['--audio-env']
+    output = args['--output']
+    paths = args['PATH']
+
+    import json
+    from functools import partial
+    import time
+    from os import environ
+    from os.path import abspath
+    from os.path import join
+    from os.path import realpath
+    from typing import List
+
+    from tqdm import tqdm
+
+    from sonusai import SonusAIError
+    from sonusai import create_file_handler
+    from sonusai import initial_log_messages
+    from sonusai import logger
+    from sonusai import update_console_handler
+    from sonusai.utils import braced_iglob
+    from sonusai.utils import p_tqdm_map
+    from sonusai.utils import seconds_to_hms
+    from sonusai.utils.asr_manifest_functions import PathInfo
+    from sonusai.utils.asr_manifest_functions import collect_librispeech_transcripts
+    from sonusai.utils.asr_manifest_functions import collect_vctk_noisy_speech_transcripts
+    from sonusai.utils.asr_manifest_functions import get_librispeech_manifest_entry
+    from sonusai.utils.asr_manifest_functions import get_vctk_noisy_speech_manifest_entry
+
+    start_time = time.monotonic()
+
+    create_file_handler('mkmanifest.log')
+    update_console_handler(verbose)
+    initial_log_messages('mkmanifest')
+
+    if method not in VALID_METHOD:
+        raise SonusAIError(f'Unknown method: {method}')
+
+    audio_dir = None
+    if audio_env is not None:
+        audio_dir = realpath(environ[audio_env])
+        if audio_dir is None:
+            raise SonusAIError(f'Unknown environment variable: {audio_env}')
+
+    if audio_env:
+        for p in paths:
+            if not realpath(abspath(p)).startswith(audio_dir):
+                logger.warning(f'Specified directory, {p}, is not part of the provided audio environment: '
+                               f'${audio_env}={audio_dir}')
+
+    logger.info('')
+    logger.info(f'Searching {len(paths)} provided director{"ies" if len(paths) > 1 else "y"}...')
+
+    entries: List[PathInfo] = []
+    for p in paths:
+        location = join(realpath(abspath(p)), '**', include)
+        logger.debug(f'Processing {location}')
+        for file in braced_iglob(pathname=location, recursive=True):
+            name = file
+            if audio_env is not None:
+                name = name.replace(audio_dir, f'${audio_env}')
+            entries.append(PathInfo(abs_path=file, audio_filepath=name))
+        logger.debug('')
+
+    logger.info(f'Found {len(entries)} audio file{"s" if len(entries) != 1 else ""}')
+
+    if dry_run:
+        logger.info('')
+        logger.info('Dry run')
+        logger.info('')
+        for entry in entries:
+            logger.info(f' - {entry.audio_filepath}')
+        return
+
+    if method == 'librispeech':
+        logger.info('Collecting LibriSpeech transcript data')
+        transcript_data = collect_librispeech_transcripts(paths=paths)
+
+        processing_func = partial(get_librispeech_manifest_entry, transcript_data=transcript_data)
+        progress = tqdm(total=len(entries), desc='Creating LibriSpeech manifest data')
+        results = p_tqdm_map(processing_func, entries, progress=progress, chunksize=10)
+        progress.close()
+
+        with open(output, 'w') as f:
+            for result in results:
+                f.write(json.dumps(result) + '\n')
+
+    if method == 'vctk_noisy_speech':
+        logger.info('Collecting VCTK Noisy Speech transcript data')
+        transcript_data = collect_vctk_noisy_speech_transcripts(paths=paths)
+
+        processing_func = partial(get_vctk_noisy_speech_manifest_entry, transcript_data=transcript_data)
+        progress = tqdm(total=len(entries), desc='Creating VCTK Noisy Speech manifest data')
+        results = p_tqdm_map(processing_func, entries, progress=progress, chunksize=10)
+        progress.close()
+
+        with open(output, 'w') as f:
+            for result in results:
+                f.write(json.dumps(result) + '\n')
+
+    end_time = time.monotonic()
+    logger.info('')
+    logger.info(f'Completed in {seconds_to_hms(seconds=end_time - start_time)}')
+    logger.info('')
+
+
+if __name__ == '__main__':
+    try:
+        main()
+    except KeyboardInterrupt:
+        logger.info('Canceled due to keyboard interrupt')
+        raise SystemExit(0)

{sonusai-0.12.6 → sonusai-0.12.7}/sonusai/mkwav.py

@@ -85,7 +85,7 @@ def _process_mixture(mixid: int) -> None:
         with h5py.File(mixture_filename, 'r') as f:
             mixture = np.array(f['mixture'])
             if MP_GLOBAL.write_target:
-                target = sum(np.array(f['targets']))
+                target = np.sum(np.array(f['targets']), axis=0)
             if MP_GLOBAL.write_noise:
                 noise = np.array(f['noise'])
 
@@ -148,7 +148,7 @@ def main() -> None:
     MP_GLOBAL.write_noise = write_noise
 
     progress = tqdm(total=len(mixid))
-    p_tqdm_map(_process_mixture, mixid, progress=progress)
+    p_tqdm_map(_process_mixture, mixid, progress=progress, chunksize=10)
     progress.close()
 
     logger.info(f'Wrote {len(mixid)} mixtures to {location}')

{sonusai-0.12.6 → sonusai-0.12.7}/sonusai/post_spenh_targetf.py

@@ -101,7 +101,7 @@ def main() -> None:
     logger.info(f'Found {len(input_name):,} files to process')
 
     progress = tqdm(total=len(input_name))
-    p_tqdm_map(_process, input_name, progress=progress)
+    p_tqdm_map(_process, input_name, progress=progress, chunksize=10)
     progress.close()
 
     logger.info(f'Wrote {len(input_name)} mixtures to {output_dir}')

{sonusai-0.12.6 → sonusai-0.12.7}/sonusai/utils/__init__.py

@@ -3,6 +3,7 @@ from sonusai.utils.asl_p56 import asl_p56
 from sonusai.utils.asr import ASRResult
 from sonusai.utils.asr import calc_asr
 from sonusai.utils.braced_glob import braced_glob
+from sonusai.utils.braced_glob import braced_iglob
 from sonusai.utils.calculate_input_shape import calculate_input_shape
 from sonusai.utils.create_ts_name import create_ts_name
 from sonusai.utils.dataclass_from_dict import dataclass_from_dict

{sonusai-0.12.6 → sonusai-0.12.7}/sonusai/utils/asr.py

@@ -1,5 +1,6 @@
 from dataclasses import dataclass
 from typing import Any
+from typing import Optional
 from typing import Union
 
 from sonusai.mixture import AudioT
@@ -9,14 +10,14 @@ from sonusai.mixture import Location
 @dataclass(frozen=True)
 class ASRResult:
     text: str
-    confidence: float = None
+    confidence: Optional[float] = None
 
 
 def calc_asr(audio: Union[AudioT, Location],
-             engine: str = 'deepgram',
-             whisper_model: Any = None,
-             whisper_model_name: str = 'base.en',
-             device: str = None) -> ASRResult:
+             engine: Optional[str] = 'deepgram',
+             whisper_model: Optional[Any] = None,
+             whisper_model_name: Optional[str] = 'base.en',
+             device: Optional[str] = None) -> ASRResult:
     """Run ASR on audio
 
     :param audio: Numpy array of audio samples or location of an audio file

{sonusai-0.12.6 → sonusai-0.12.7}/sonusai/utils/asr_functions/data.py

@@ -1,5 +1,6 @@
 from dataclasses import dataclass
 from typing import Any
+from typing import Optional
 
 from sonusai.mixture.types import AudioT
 
@@ -7,6 +8,6 @@ from sonusai.mixture.types import AudioT
 @dataclass(frozen=True)
 class Data:
     audio: AudioT
-    whisper_model: Any = None
-    whisper_model_name: str = None
-    device: str = None
+    whisper_model: Optional[Any] = None
+    whisper_model_name: Optional[str] = None
+    device: Optional[str] = None

sonusai-0.12.7/sonusai/utils/asr_manifest_functions/__init__.py

@@ -0,0 +1,6 @@
+from sonusai.utils.asr_manifest_functions.data import PathInfo
+from sonusai.utils.asr_manifest_functions.data import TranscriptData
+from sonusai.utils.asr_manifest_functions.librispeech import collect_librispeech_transcripts
+from sonusai.utils.asr_manifest_functions.librispeech import get_librispeech_manifest_entry
+from sonusai.utils.asr_manifest_functions.vctk_noisy_speech import collect_vctk_noisy_speech_transcripts
+from sonusai.utils.asr_manifest_functions.vctk_noisy_speech import get_vctk_noisy_speech_manifest_entry

sonusai-0.12.7/sonusai/utils/asr_manifest_functions/data.py

@@ -0,0 +1,10 @@
+from dataclasses import dataclass
+from typing import Dict
+
+TranscriptData = Dict[str, str]
+
+
+@dataclass(frozen=True)
+class PathInfo:
+    abs_path: str
+    audio_filepath: str

sonusai-0.12.7/sonusai/utils/asr_manifest_functions/librispeech.py

@@ -0,0 +1,49 @@
+from typing import List
+from typing import Union
+
+from sonusai.utils.asr_manifest_functions import PathInfo
+from sonusai.utils.asr_manifest_functions import TranscriptData
+
+
+def collect_librispeech_transcripts(paths: Union[List[str], str]) -> TranscriptData:
+    from glob import iglob
+    from os.path import abspath
+    from os.path import dirname
+    from os.path import join
+
+    from sonusai import SonusAIError
+
+    entries: TranscriptData = {}
+    if not isinstance(paths, list):
+        paths = [paths]
+
+    for p in paths:
+        location = join(abspath(p), '**', '*.trans.txt')
+        for file in iglob(pathname=location, recursive=True):
+            root = dirname(file)
+            with open(file, encoding='utf-8') as f:
+                for line in f:
+                    name, text = line[: line.index(' ')], line[line.index(' ') + 1:]
+                    name = join(root, name)
+                    if name in entries:
+                        raise SonusAIError(f'{name} already exists in transcript data')
+                    entries[name] = text.lower().strip()
+    return entries
+
+
+def get_librispeech_manifest_entry(entry: PathInfo, transcript_data: TranscriptData) -> dict:
+    from os.path import splitext
+    from subprocess import check_output
+
+    from sonusai import SonusAIError
+
+    name = splitext(entry.abs_path)[0]
+    duration = float(check_output(f'soxi -D {entry.abs_path}', shell=True))
+    if name not in transcript_data.keys():
+        raise SonusAIError(f'Could not find {name} in transcript data')
+
+    return {
+        'audio_filepath': entry.audio_filepath,
+        'text':           transcript_data[name],
+        'duration':       duration,
+    }

sonusai-0.12.7/sonusai/utils/asr_manifest_functions/vctk_noisy_speech.py

@@ -0,0 +1,69 @@
+from typing import List
+from typing import Union
+
+from sonusai.utils.asr_manifest_functions import PathInfo
+from sonusai.utils.asr_manifest_functions import TranscriptData
+
+
+def collect_vctk_noisy_speech_transcripts(paths: Union[List[str], str]) -> TranscriptData:
+    from glob import iglob
+    from os import listdir
+    from os.path import abspath
+    from os.path import basename
+    from os.path import join
+    from os.path import split
+    from os.path import splitext
+
+    from sonusai import SonusAIError
+
+    entries: TranscriptData = {}
+    if not isinstance(paths, list):
+        paths = [paths]
+
+    for p in paths:
+        abs_p = abspath(p)
+        head, tail = split(abs_p)
+
+        dirs = listdir(head)
+        tail = tail.replace('wav', 'txt')
+
+        location = None
+        for d in dirs:
+            if tail.endswith(d):
+                location = join(head, d, '*.txt')
+                break
+        if location is None:
+            raise SonusAIError(f'Could not find VCTK Noisy Speech transcript data for {p}')
+
+        for file in iglob(pathname=location, recursive=True):
+            with open(file, encoding='utf-8') as f:
+                lines = f.readlines()
+                if len(lines) != 1:
+                    raise SonusAIError(f'Ill-formed VCTK Noisy Speech transcript file: {file}')
+
+                name = join(abs_p, splitext(basename(file))[0])
+                text = lines[0].lower().strip()
+
+                if name in entries:
+                    raise SonusAIError(f'{name} already exists in transcript data')
+                entries[name] = text.lower().strip()
+
+    return entries
+
+
+def get_vctk_noisy_speech_manifest_entry(entry: PathInfo, transcript_data: TranscriptData) -> dict:
+    from os.path import splitext
+    from subprocess import check_output
+
+    from sonusai import SonusAIError
+
+    name = splitext(entry.abs_path)[0]
+    duration = float(check_output(f'soxi -D {entry.abs_path}', shell=True))
+    if name not in transcript_data.keys():
+        raise SonusAIError(f'Could not find {name} in transcript data')
+
+    return {
+        'audio_filepath': entry.audio_filepath,
+        'text':           transcript_data[name],
+        'duration':       duration,
+    }

{sonusai-0.12.6 → sonusai-0.12.7}/sonusai/utils/braced_glob.py

@@ -32,11 +32,18 @@ def expand_braces(text: str, seen: Optional[Set[str]] = None) -> Generator[str,
             yield from expand_braces(''.join(replaced), seen)
 
 
-def braced_glob(path: str) -> List[str]:
+def braced_glob(pathname: str, recursive: bool = False) -> List[str]:
     from glob import glob
 
     result = []
-    for x in expand_braces(path):
-        result.extend(glob(x))
+    for expanded_path in expand_braces(pathname):
+        result.extend(glob(expanded_path, recursive=recursive))
 
     return result
+
+
+def braced_iglob(pathname: str, recursive: bool = False) -> Generator[str, None, None]:
+    from glob import iglob
+
+    for expanded_path in expand_braces(pathname):
+        yield from iglob(expanded_path, recursive=recursive)

{sonusai-0.12.6 → sonusai-0.12.7}/sonusai/utils/parallel_tqdm.py

@@ -11,7 +11,7 @@ from typing import Iterable
 from typing import List
 
 
-def _parallel(ordered: bool, function: Callable, *iterables: Iterable, **kwargs: Any) -> Generator:
+def __parallel(ordered: bool, function: Callable, *iterables: Iterable, **kwargs: Any) -> Generator:
     """Returns a generator for a parallel map.
 
     Arguments:
@@ -43,6 +43,7 @@ def _parallel(ordered: bool, function: Callable, *iterables: Iterable, **kwargs:
 
     # Extract num_cpus
     num_cpus = kwargs.pop('num_cpus', None)
+    chunksize = kwargs.pop('chunksize', 1)
 
     # Determine num_cpus
     if num_cpus is None:
@@ -57,7 +58,7 @@ def _parallel(ordered: bool, function: Callable, *iterables: Iterable, **kwargs:
     with mp.Pool(num_cpus, initializer=initializer, initargs=initargs) as pool:
         map_func = getattr(pool, map_type)
 
-        for item in map_func(function, *iterables):
+        for item in map_func(function, *iterables, chunksize=chunksize):
             yield item
             progress.update()
 
@@ -67,9 +68,9 @@ def _parallel(ordered: bool, function: Callable, *iterables: Iterable, **kwargs:
 
 def p_tqdm_map(function: Callable, *iterables: Iterable, **kwargs: Any) -> List[Any]:
     """Performs a parallel ordered map."""
-    return list(_parallel(True, function, *iterables, **kwargs))
+    return list(__parallel(True, function, *iterables, **kwargs))
 
 
 def p_tqdm_umap(function: Callable, *iterables: Iterable, **kwargs: Any) -> List[Any]:
     """Performs a parallel unordered map."""
-    return list(_parallel(False, function, *iterables, **kwargs))
+    return list(__parallel(False, function, *iterables, **kwargs))