[FEATURE] Audio normalization

docs/source/api.rst

@@ -5,4 +5,5 @@
    :maxdepth: 2
 
    publicapi
-   coreapi
+   coreapi
+   utils

docs/source/coreapi_usage.rst

@@ -110,6 +110,59 @@ The data is fetched seamlessly on-demand from the audio file(s). The opening/clo
 
 Eventual time gap between audio items are filled with ``0.`` values.
 
+Normalization
+"""""""""""""
+
+The fetched audio data can be normalized according to the presets given by the :class:`osekit.utils.audio_utils.Normalization` flag:
+
+.. list-table:: Normalization presets
+   :widths: 10 10
+   :header-rows: 1
+
+   * - Name
+     - Description
+   * - ``Normalization.RAW``
+     - :math:`x`
+   * - ``Normalization.DC_REJECT``
+     - :math:`x-\overline{ x }`
+   * - ``Normalization.PEAK``
+     - :math:`\frac{x}{x_\text{max}}`
+   * - ``Normalization.ZSCORE``
+     - :math:`\frac{ x-\overline{x} }{\sigma (x)}`
+
+To normalize the data, simply set the :attr:`osekit.core_api.audio_data.AudioData.normalization` property to the
+requested normalization flag:
+
+.. code-block:: python
+
+    from osekit.core_api.audio_data.AudioData import AudioData
+    from osekit.utils.audio_utils.normalization import Normalization
+
+    ad = AudioData(...)
+    ad.normalization = Normalization.ZSCORE # Note: normalization also is a parameter of the AudioData initializer
+
+    v = ad.get_value() # The fetched data will then be normalized
+
+.. note::
+
+    The ``Normalization.DC_REJECT`` normalization can be combined with any single other normalization:
+
+    .. code-block:: python
+
+        from osekit.utils.audio_utils.normalization import Normalization
+
+        dc_peak = Normalization.DC_REJECT | Normalization.PEAK
+
+.. warning::
+
+    Instantiating another combination of normalizations will raise an error:
+
+    .. code-block:: python
+
+        from osekit.utils.audio_utils.normalization import Normalization
+
+        incorrect_normalization = Normalization.RAW | Normalization.PEAK
+        incorrect_normalization = Normalization.DC_REJECT | Normalization.RAW | Normalization.PEAK
 
 Calibration
 """""""""""
@@ -124,8 +177,8 @@ allows for retrieving the data in the shape of the recorded acoustic pressure.
 
 .. code-block:: python
 
-    from osekit.core_api.instrument import Instrument
     from osekit.core_api.audio_data import AudioData
+    from osekit.core_api.instrument import Instrument
     import numpy as np
 
     instrument = Instrument(end_to_end_db = 150) # The raw 1. WAV value equals 150 dB SPL re 1 uPa
@@ -170,6 +223,7 @@ an ``AudioDataset`` from a given folder containing audio files:
 
     from pathlib import Path
     from osekit.core_api.audio_dataset import AudioDataset
+    from osekit.core_api.instrument import Instrument
     from pandas import Timestamp, Timedelta
 
     folder = Path(r"...")
@@ -179,7 +233,9 @@ an ``AudioDataset`` from a given folder containing audio files:
         strptime_format="%y_%m_%d_%H_%M_%S", # To parse the files begin Timestamp
         begin=Timestamp("2009-01-06 12:00:00"),
         end=Timestamp("2009-01-06 14:00:00"),
-        data_duration=Timedelta("10s")
+        data_duration=Timedelta("10s"),
+        instrument=Instrument(end_to_end_db=150),
+        normalization="dc_reject"
     )
 
 The resulting ``AudioDataset`` will contain 10s-long ``AudioData`` ranging from ``2009-01-06 12:00:00`` to ``2009-01-06 14:00:00``.
@@ -366,4 +422,4 @@ should be provided:
     ltas.plot()
     plt.show()
 
-A ``SpectroData`` object can be turned into a ``LTASData`` thanks to the :meth:`osekit.core_api.ltas_data.LTASData.from_spectro_data` method.
+A ``SpectroData`` object can be turned into a ``LTASData`` thanks to the :meth:`osekit.core_api.ltas_data.LTASData.from_spectro_data` method.

docs/source/example_ltas.rst

@@ -13,6 +13,7 @@ This LTAS will:
 * Start at the begin of the first audio file
 * End at the end of the last audio file
 * Be downsampled at ``24 kHz``
+* Have its DC component removed
 
 | The FFT used for computing the spectrograms will use a ``1024 samples``-long hamming window.
 | The ``hop`` of LTAS ``ShortTimeFFT`` objects is forced to the size of the window (no overlap).

docs/source/example_ltas_core.ipynb

@@ -51,6 +51,7 @@
     "audio_folder = Path(r\"_static/sample_audio\")\n",
     "\n",
     "from osekit.core_api.audio_dataset import AudioDataset\n",
+    "from osekit.utils.audio_utils import Normalization\n",
     "from osekit.core_api.instrument import Instrument\n",
     "\n",
     "audio_data = AudioDataset.from_folder(\n",
@@ -60,7 +61,10 @@
     ").data[0]\n",
     "\n",
     "# Resampling at 24 kHz\n",
-    "audio_data.sample_rate = 24_000"
+    "audio_data.sample_rate = 24_000\n",
+    "\n",
+    "# Removing the DC component\n",
+    "audio_data.normalization = Normalization.DC_REJECT"
    ]
   },
   {

docs/source/example_ltas_public.ipynb

@@ -134,13 +134,15 @@
    "metadata": {},
    "outputs": [],
    "source": [
+    "from osekit.utils.audio_utils import Normalization\n",
     "from osekit.public_api.analysis import Analysis, AnalysisType\n",
     "\n",
     "analysis = Analysis(\n",
     "    analysis_type=AnalysisType.SPECTROGRAM\n",
     "    | AnalysisType.MATRIX,  # we want to export both the spectrogram and the sx matrix\n",
     "    nb_ltas_time_bins=3000,  # This will turn the regular spectrum computation in a LTAS\n",
     "    sample_rate=sample_rate,\n",
+    "    normalization=Normalization.DC_REJECT,  # Removes the DC component\n",
     "    fft=sft,\n",
     "    v_lim=(0.0, 150.0),  # Boundaries of the spectrograms\n",
     "    colormap=\"viridis\",  # Default value\n",
@@ -196,7 +198,11 @@
    "cell_type": "code",
    "execution_count": null,
    "id": "e05d653bc1e8bfe2",
-   "metadata": {},
+   "metadata": {
+    "tags": [
+     "remove-cell"
+    ]
+   },
    "outputs": [],
    "source": [
     "# Reset the dataset to get all files back to place.\n",

docs/source/example_multiple_spectrograms.rst

@@ -9,6 +9,7 @@ In this example, we want to export spectrograms drawn from the sample audio data
 * Last spectrogram ends at ``2022-09-25 22:36:25``
 * Spectrograms represent ``5 s``-long audio data
 * Audio data are downsampled sampled at ``24 kHz`` before spectrograms are computed
+* The DC component of the audio data is rejected before spectrograms are computed
 * Spectrograms that are in the gap between recordings should be skipped
 
 The FFT used for computing the spectrograms will use a ``1024 samples``-long hamming window, with a ``128 samples``-long hop.

docs/source/example_multiple_spectrograms_core.ipynb

@@ -52,6 +52,7 @@
     "\n",
     "from osekit.core_api.audio_dataset import AudioDataset\n",
     "from osekit.core_api.instrument import Instrument\n",
+    "from osekit.utils.audio_utils import Normalization\n",
     "from pandas import Timestamp, Timedelta\n",
     "\n",
     "audio_dataset = AudioDataset.from_folder(\n",
@@ -61,6 +62,8 @@
     "    end=Timestamp(\"2022-09-25 22:36:25\"),\n",
     "    data_duration=Timedelta(seconds=5),\n",
     "    instrument=Instrument(end_to_end_db=150.0),\n",
+    "    sample_rate=24_000,\n",
+    "    normalization=Normalization.DC_REJECT,\n",
     ")"
    ]
   },
@@ -192,7 +195,7 @@
    "source": [
     "import matplotlib.pyplot as plt\n",
     "\n",
-    "spectro_dataset.data[0].plot()\n",
+    "spectro_dataset.data[1].plot()\n",
     "plt.show()"
    ]
   },

docs/source/example_multiple_spectrograms_public.ipynb

@@ -133,6 +133,7 @@
    "outputs": [],
    "source": [
     "from osekit.public_api.analysis import Analysis, AnalysisType\n",
+    "from osekit.utils.audio_utils import Normalization\n",
     "from pandas import Timestamp, Timedelta\n",
     "\n",
     "analysis = Analysis(\n",
@@ -143,6 +144,7 @@
     "    end=Timestamp(\"2022-09-25 22:36:25\"),\n",
     "    data_duration=Timedelta(seconds=5),\n",
     "    sample_rate=sample_rate,\n",
+    "    normalization=Normalization.DC_REJECT,\n",
     "    fft=sft,\n",
     "    v_lim=(0.0, 150.0),  # Boundaries of the spectrograms\n",
     "    colormap=\"viridis\",  # Default value\n",

docs/source/example_reshaping_multiple_files.rst

@@ -9,6 +9,7 @@ In this example, we want to export reshaped files from the sample audio dataset
 * Last file ends at ``2022-09-25 22:36:25``
 * Files are ``5 s``-long
 * Files are sampled at ``24 kHz``
+* Files are DC-filtered
 * Files that are in the gap between recordings should be skipped
 
 .. toctree::

docs/source/example_reshaping_multiple_files_core.ipynb

@@ -47,6 +47,7 @@
     "audio_folder = Path(r\"_static/sample_audio\")\n",
     "\n",
     "from osekit.core_api.audio_dataset import AudioDataset\n",
+    "from osekit.utils.audio_utils import Normalization\n",
     "from pandas import Timestamp, Timedelta\n",
     "\n",
     "audio_dataset = AudioDataset.from_folder(\n",
@@ -55,6 +56,8 @@
     "    begin=Timestamp(\"2022-09-25 22:35:15\"),\n",
     "    end=Timestamp(\"2022-09-25 22:36:25\"),\n",
     "    data_duration=Timedelta(seconds=5),\n",
+    "    sample_rate=24_000,\n",
+    "    normalization=Normalization.DC_REJECT,\n",
     ")"
    ],
    "outputs": [],