Violin pitch extraction and few improvements (#98)

* kwargs for load model, violin pitch

* wrongly added model files :)

* set max numpy versions

* remove sample rate from diffsep config

* several fixes, mini-bug, and typos around

* use proper arg names for essentia melodia

* fix issues in READMEs, add violin model to docs

* sampleRate-->sample_rate in deepsrgm

* sampleRate-->sample_rate again...

.gitignore

@@ -73,6 +73,9 @@ compiam/models/structure/dhrupad_bandish_segmentation/annotations/section_bounda
 compiam/models/structure/dhrupad_bandish_segmentation/annotations/cycle_boundaries/
 tests/resources/mir_datasets/*
 compiam/models/separation/
+compiam/models/structure/
+compiam/models/melody/
+compiam/models/rhythm/
 
 # For next release
 compiam/models/rhythm/4wayTabla/

README.md

@@ -94,7 +94,7 @@ Please check the [contribution guidelines](https://mtg.github.io/compIAM/source/
 We include, in this repo, [example notebooks](https://github.com/MTG/compIAM/tree/master/notebooks) for users to better understand how to use `compiam` and also showcase 
 
 ## License
-compIAM is Copyright 2023 Music Technology Group - Universitat Pompeu Fabra
+compIAM is Copyright 2024 Music Technology Group - Universitat Pompeu Fabra
 
 compIAM is released under the terms of the GNU Affero General Public License (v3 or later). See the COPYING file for more information. For the case of a particular tool or implementation that has a specific different licence, this is explicitly specified in the files related to this tool, and these terms must be followed.
 

compiam/__init__.py

@@ -9,7 +9,7 @@
 from compiam.exceptions import ModelNotDefinedError
 
 
-def load_model(model_name, data_home=None, models_dict=models_dict):
+def load_model(model_name, data_home=None, models_dict=models_dict, **kwargs):
     """Wrapper for loading pre-trained models.
 
     :param model_name: name of the model, extractors, or algorithm to load.
@@ -41,6 +41,7 @@ def load_model(model_name, data_home=None, models_dict=models_dict):
             )
 
     module = getattr(import_module(m_dict["module_name"]), m_dict["class_name"])
+    m_dict["kwargs"].update(kwargs)
     return module(**m_dict["kwargs"])
 
 
@@ -103,7 +104,7 @@ def get_model_info(model_key):
     """Get complete info in data/models_dict for a particular pre-trained model
 
     :param model_key: model key from models_dict
-    :returns: infomation about a particular model.
+    :returns: information about a particular model.
     """
     if model_key not in list(models_dict.keys()):
         raise ValueError(

compiam/data.py

@@ -52,6 +52,16 @@
             "sample_rate": 8000,
         },
     },
+    "melody:ftaresnet-carnatic-violin": {
+        "module_name": "compiam.melody.pitch_extraction.ftaresnet_carnatic",
+        "class_name": "FTAResNetCarnatic",
+        "kwargs": {
+            "model_path": os.path.join(
+                "models", "melody", "ftanet", "fta_carnatic_violin", "FTA-ResNet_best_version.pth"
+            ),
+            "sample_rate": 44100,
+        },
+    },
     "melody:cae-carnatic": {
         "module_name": "compiam.melody.pattern.sancara_search",
         "class_name": "CAEWrapper",
@@ -114,7 +124,7 @@
                 "saraga-8",
                 "saraga-8.ckpt-1",
             ),
-            "sample_rate": 22050,
+            #"sample_rate": 22050,  # Already contained in the model config
         },
     },
 }

compiam/dunya/README.md

@@ -2,4 +2,4 @@
 
 Built within the framework of the [CompMusic project](https://compmusic.upf.edu/), the Dunya corpora cover several World music traditions and can be accessed through different ways. In compIAM we provide tools to easily access the Carnatic and Hindustani corpora in Dunya.
 
-When initializing the corpora access through compIAM you will be asked to provide an *access token*. To get the *access token* you need to register to the [Dunya web portal](https://dunya.compmusic.upf.edu/). Once your registration is validation by the CompMusic team, you can access your user account information where an access token is provided. Use this token to intialize the ``compiam.Corpora`` objects and access the corpora.
+When initializing the corpora access through compIAM you will be asked to provide an *access token*. To get the *access token* you need to register to the [Dunya web portal](https://dunya.compmusic.upf.edu/). Once your registration is validation by the CompMusic team, you can access your user account information where an access token is provided. Use this token to initialize the ``compiam.Corpora`` objects and access the corpora.

compiam/dunya/__init__.py

@@ -312,7 +312,7 @@ def save_annotation(
             )
 
     def download_mp3(self, recording_id, output_dir):
-        """Download the mp3 of a document and save it to the specificed directory.
+        """Download the mp3 of a document and save it to the specified directory.
 
         :param recording_id: The MBID of the recording.
         :param output_dir: Where to save the mp3 to.
@@ -338,7 +338,7 @@ def download_mp3(self, recording_id, output_dir):
         return name
 
     def download_concert(self, concert_id, output_dir):
-        """Download the mp3s of all recordings in a concert and save them to the specificed directory.
+        """Download the mp3s of all recordings in a concert and save them to the specified directory.
 
         :param concert_id: The MBID of the concert.
         :param location: Where to save the mp3s to.

compiam/dunya/conn.py

@@ -54,6 +54,7 @@ def set_token(token):
 
 
 def _dunya_url_query(url, extra_headers=None):
+    """Use requests to query from a given url in Dunya."""
     logger.debug("query to '%s'" % url)
     if not TOKEN:
         raise ConnectionError("You need to authenticate with `set_token`")
@@ -71,6 +72,7 @@ def _dunya_url_query(url, extra_headers=None):
 
 
 def _make_url(path, **kwargs):
+    """Create URL from path."""
     if "://" in HOSTNAME:
         protocol, hostname = HOSTNAME.split("://")
     else:

compiam/io.py

@@ -38,14 +38,18 @@ def read_csv(file_path):
     return output[~np.isnan(output)]
 
 
-def save_object(obj, filename):
-    with open(filename, "wb") as outp:  # Overwrites any existing file.
+def save_object(obj, file_path):
+    """Saves object to pickle file
+    
+    :param obj: an object to save to pickle file
+    :param file_path: path to save the object
+    """
+    with open(file_path, "wb") as outp:  # Overwrites any existing file.
         pickle.dump(obj, outp, pickle.HIGHEST_PROTOCOL)
 
 
 def write_json(j, path):
-    """
-    Write json, <j>, to <path>
+    """Write json, <j>, to <path>
 
     :param j: json
     :type path: json

compiam/melody/README.md

@@ -4,16 +4,18 @@
 |---------------------------|----------------------------------|-----------|
 | Melodia                   | Predominant pitch extraction     | [1]       |
 | FTANet-Carnatic           | Vocal pitch extraction           | [2]       |
-| TonicIndianArtMusic       | Tonic idendification             | [3]       |
-| CAE + Matrix profile      | Melodic pattern discovery        | [4]       |
-| DEEPSRGM                  | DL-based raga recognition        | [5]       |
+| FTAResNet-Carnatic-Violin | Violin pitch extraction          | [3]       |
+| TonicIndianArtMusic       | Tonic idendification             | [4]       |
+| CAE + Matrix profile      | Melodic pattern discovery        | [5]       |
+| DEEPSRGM                  | DL-based raga recognition        | [6]       |
+
 
 [1] J. Salamon and E. Gómez, "Melody extraction from polyphonic music signals using pitch contour characteristics", IEEE Transactions on Audio, Speech, and Language Processing, vol. 20, no. 6, pp. 1759–1770, 2012.
 
-[2] G. Plaja-Roglans, T. Nuttall, L. Pearson, X. Serra, M. Miron, "Repertoire-Specific Vocal Pitch Data Generation for Improved Melodic Analysis of Carnatic Music", Transactions of the International Society for Music Information Retrieval, vol. 6, no. 1, pp. 13–26, 2023.
+[2, 3] G. Plaja-Roglans, T. Nuttall, L. Pearson, X. Serra, M. Miron, "Repertoire-Specific Vocal Pitch Data Generation for Improved Melodic Analysis of Carnatic Music", Transactions of the International Society for Music Information Retrieval, vol. 6, no. 1, pp. 13–26, 2023.
 
-[3] J. Salamon, S. Gulati, and X. Serra, "A Multipitch Approach to Tonic Identification in Indian Classical Music", in International Society for Music Information Retrieval Conference (ISMIR 12), 2012. 
+[4] J. Salamon, S. Gulati, and X. Serra, "A Multipitch Approach to Tonic Identification in Indian Classical Music", in International Society for Music Information Retrieval Conference (ISMIR 12), 2012. 
 
-[4] T. Nuttall, G. Plaja-Roglans, L. Pearson, and X. Serra, "In search of sañcaras: tradition-informed repeated melodic pattern recognition in Carnatic Music", in International Society for Music Information Retrieval Conference (ISMIR 22), 2022. 
+[5] T. Nuttall, G. Plaja-Roglans, L. Pearson, and X. Serra, "In search of sañcaras: tradition-informed repeated melodic pattern recognition in Carnatic Music", in International Society for Music Information Retrieval Conference (ISMIR 22), 2022. 
 
-[5] S. Madhusudhan and G. Chowdhary, "DeepSRGM - Sequence Classification and Ranking in Indian Classical Music Via Deep Learning", in International Society for Music Information Retrieval Conference (ISMIR 20), 2020.
+[6] S. Madhusudhan and G. Chowdhary, "DeepSRGM - Sequence Classification and Ranking in Indian Classical Music Via Deep Learning", in International Society for Music Information Retrieval Conference (ISMIR 20), 2020.

compiam/melody/pattern/sancara_search/complex_auto/complex.py

@@ -13,6 +13,7 @@
 
 
 class Complex(nn.Module):
+    """CAE model."""
     def __init__(self, n_in, n_out, dropout=0.5, learn_norm=False):
         super(Complex, self).__init__()
 

compiam/melody/pitch_extraction/__init__.py

@@ -5,6 +5,7 @@
 
 from compiam.melody.pitch_extraction.melodia import Melodia
 from compiam.melody.pitch_extraction.ftanet_carnatic import FTANetCarnatic
+from compiam.melody.pitch_extraction.ftaresnet_carnatic import FTAResNetCarnatic
 
 
 # Show user the available tools

compiam/melody/pitch_extraction/ftanet_carnatic/__init__.py

@@ -27,6 +27,7 @@ def __init__(self, model_path=None, sample_rate=8000, gpu="-1"):
 
         :param model_path: path to file to the model weights.
         :param sample_rate: Sample rate to which the audio is sampled for extraction.
+        :param gpu: Id of the available GPU to use (-1 by default, to run on CPU), use string: '0', '1', etc.
         """
         ### IMPORTING OPTIONAL DEPENDENCIES
         try:
@@ -276,7 +277,7 @@ def predict(
             (defaulted to 5, increase if enough computational power, reduce if
             needed).
         :param out_step: particular time-step duration if needed at output
-        :param gpu: Id of the available GPU to use (-1 by default, to run on CPU)
+        :param gpu: Id of the available GPU to use (-1 by default, to run on CPU), use string: '0', '1', etc.
         :returns: a 2-D list with time-stamps and pitch values per timestamp.
         """
         ## Setting up GPU if any

compiam/melody/pitch_extraction/ftanet_carnatic/pitch_processing.py

@@ -112,3 +112,18 @@ def get_est_arr(model, x_list, y_list, batch_size):
         est_arr = est(preds, CenFreq, y)
 
     return est_arr
+
+
+def std_normalize(data): 
+    """Standardize the input data.
+    
+    :param data: input data.
+    :returns: standardized data.
+    """
+    data = data.astype(np.float64)
+    mean = np.mean(data)
+    std = np.std(data)
+    data = data.copy() - mean
+    if std != 0.:
+        data = data / std
+    return data.astype(np.float32)