nkululeko 0.92.2__py3-none-any.whl → 0.93.0__py3-none-any.whl

nkululeko/autopredict/ap_sid.py

@@ -6,8 +6,11 @@ import numpy as np
 from pyannote.audio import Pipeline
 import torch
 
+import audiofile
+
 from nkululeko.feature_extractor import FeatureExtractor
 import nkululeko.glob_conf as glob_conf
+from nkululeko.utils.files import concat_files
 from nkululeko.utils.util import Util
 
 
@@ -20,7 +23,7 @@ class SIDPredictor:
     def __init__(self, df):
         self.df = df
         self.util = Util("sidPredictor")
-        hf_token = self.util.config_val("Model", "hf_token", None)
+        hf_token = self.util.config_val("MODEL", "hf_token", None)
         if hf_token is None:
             self.util.error(
                 "speaker id prediction needs huggingface token: [MODEL][hf_token]"
@@ -29,20 +32,45 @@ class SIDPredictor:
             "pyannote/speaker-diarization-3.1",
             use_auth_token=hf_token,
         )
-        device = self.util.config_val("Model", "device", "cpu")
+        device = self.util.config_val("MODEL", "device", "cpu")
         self.pipeline.to(torch.device(device))
 
     def predict(self, split_selection):
         self.util.debug(f"estimating speaker id for {split_selection} samples")
         return_df = self.df.copy()
-        # @todo
         # 1) concat all audio files
+        tmp_file = "tmp.wav"
+        concat_files(return_df.index, tmp_file)
         # 2) get segmentations with pyannote
-        # 3) map pyannote segments with orginal ones and assign speaker id
+        sname = "pyannotation"
+        if self.util.exist_pickle(sname):
+            annotation = self.util.from_pickle(sname)
+        else:
+            annotation = self.pipeline(tmp_file)
+            self.util.to_pickle(annotation, sname)
 
+        speakers, starts, ends = [], [], []
+        # print the result
+        for turn, _, speaker in annotation.itertracks(yield_label=True):
+            start = turn.start
+            end = turn.end
+            speakers.append(speaker)
+            starts.append(start)
+            ends.append(end)
+        # 3) map pyannote segments with orginal ones and assign speaker id
+        target_speakers = []
+        position = 0
+        for idx, (file, start, end) in enumerate(return_df.index.to_list()):
+            seg_start = start.total_seconds()
+            seg_end = end.total_seconds()
+            # file_duration = audiofile.duration(file)
+            seg_duration = seg_end - seg_start
+            offset = position + seg_start + seg_duration / 2
+            l = [i < offset for i in starts]
+            r = [i for i, x in enumerate(l) if x]
+            s_index = r.pop()
+            # self.util.debug(f"offset: {offset}, speaker = {speakers[s_index]}")
+            position += seg_duration
+            target_speakers.append(speakers[s_index])
+        return_df["speaker"] = target_speakers
         return return_df
-
-    def concat_files(self, df):
-        pass
-        # todo
-        # please use https://audeering.github.io/audiofile/usage.html#read-a-file

nkululeko/constants.py

@@ -1,2 +1,2 @@
-VERSION="0.92.2"
+VERSION="0.93.0"
 SAMPLING_RATE = 16000

nkululeko/data/dataset_csv.py

@@ -53,7 +53,7 @@ class Dataset_CSV(Dataset):
             if audformat.index_type(df.index) == "segmented":
                 file_index = (
                     df.index.levels[0]
-                    .map(lambda x: root + "/" + audio_path + "/" + x)
+                    .map(lambda x: os.path.join(root, audio_path, x))
                     .values
                 )
                 df = df.set_index(df.index.set_levels(file_index, level="file"))
@@ -62,20 +62,20 @@ class Dataset_CSV(Dataset):
                     df = pd.DataFrame(df)
                 df = df.set_index(
                     df.index.to_series().apply(
-                        lambda x: root + "/" + audio_path + "/" + x
+                        lambda x: os.path.join(root, audio_path, x)
                     )
                 )
         else:  # absolute path is True
             if audformat.index_type(df.index) == "segmented":
                 file_index = (
-                    df.index.levels[0].map(lambda x: audio_path + "/" + x).values
+                    df.index.levels[0].map(lambda x: os.path.join(audio_path, x)).values
                 )
                 df = df.set_index(df.index.set_levels(file_index, level="file"))
             else:
                 if not isinstance(df, pd.DataFrame):
                     df = pd.DataFrame(df)
                 df = df.set_index(
-                    df.index.to_series().apply(lambda x: audio_path + "/" + x)
+                    df.index.to_series().apply(lambda x: os.path.join(audio_path, x))
                 )
 
         self.df = df

nkululeko/experiment.py

@@ -422,9 +422,7 @@ class Experiment:
         return df_ret
 
     def autopredict(self):
-        """
-        Predict labels for samples with existing models and add to the dataframe.
-        """
+        """Predict labels for samples with existing models and add to the dataframe."""
         sample_selection = self.util.config_val("PREDICT", "split", "all")
         if sample_selection == "all":
             df = pd.concat([self.df_train, self.df_test])

nkululeko/feat_extract/feats_wav2vec2.py

@@ -1,4 +1,4 @@
-""" feats_wav2vec2.py
+"""feats_wav2vec2.py.
 feat_types example = [wav2vec2-large-robust-ft-swbd-300h,
 wav2vec2-xls-r-2b, wav2vec2-large, wav2vec2-large-xlsr-53, wav2vec2-base]
 
@@ -11,12 +11,13 @@ import os
 import pandas as pd
 import torch
 import torchaudio
-import transformers
 from tqdm import tqdm
-from transformers import Wav2Vec2FeatureExtractor, Wav2Vec2Model
+import transformers
+from transformers import Wav2Vec2FeatureExtractor
+from transformers import Wav2Vec2Model
 
-import nkululeko.glob_conf as glob_conf
 from nkululeko.feat_extract.featureset import Featureset
+import nkululeko.glob_conf as glob_conf
 
 
 class Wav2vec2(Featureset):

nkululeko/feat_extract/feats_wavlm.py

@@ -8,10 +8,11 @@ import pandas as pd
 import torch
 import torchaudio
 from tqdm import tqdm
-from transformers import Wav2Vec2FeatureExtractor, WavLMModel
+from transformers import Wav2Vec2FeatureExtractor
+from transformers import WavLMModel
 
-import nkululeko.glob_conf as glob_conf
 from nkululeko.feat_extract.featureset import Featureset
+import nkululeko.glob_conf as glob_conf
 
 
 class Wavlm(Featureset):

nkululeko/utils/files.py

@@ -1,6 +1,6 @@
 #!/usr/bin/env python3
 # -*- coding: utf-8 -*-
-# copied from librosa.util.files.py
+# find_files copied from librosa.util.files.py
 
 """Utility functions for dealing with files"""
 from __future__ import annotations
@@ -8,11 +8,22 @@ from __future__ import annotations
 import glob
 import os
 from pathlib import Path
-from typing import Any, List, Optional, Set, Union
+from typing import Any
+from typing import List
+from typing import Optional
+from typing import Set
+from typing import Union
+
+import numpy as np
+from tqdm import tqdm
+
+import audiofile
+
 
 # add new function here
 __all__ = [
     "find_files",
+    "concat_files",
 ]
 
 
@@ -143,3 +154,16 @@ def __get_files(dir_name: Union[str, os.PathLike[Any]], extensions: Set[str]):
         myfiles |= set(glob.glob(globstr))
 
     return myfiles
+
+
+def concat_files(index, outfile_path):
+    buffer = np.asarray([])
+    sr = 16000
+    for idx, (file, start, end) in enumerate(tqdm(index.to_list())):
+        signal, sr = audiofile.read(
+            file,
+            offset=start.total_seconds(),
+            duration=(end - start).total_seconds(),
+        )
+        buffer = np.concatenate([buffer, signal])
+    audiofile.write(outfile_path, buffer, sr)

{nkululeko-0.92.2.dist-info → nkululeko-0.93.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: nkululeko
-Version: 0.92.2
+Version: 0.93.0
 Summary: Machine learning audio prediction experiments based on templates
 Home-page: https://github.com/felixbur/nkululeko
 Author: Felix Burkhardt
@@ -355,6 +355,10 @@ F. Burkhardt, Johannes Wagner, Hagen Wierstorf, Florian Eyben and Björn Schulle
 Changelog
 =========
 
+Version 0.93.0
+--------------
+* integrated pyannote for speaker prediction for predict module
+
 Version 0.92.2
 --------------
 * added some output to automatic speaker id
@@ -365,7 +369,7 @@ Version 0.92.1
 
 Version 0.92.0
 --------------
-* added first version of automatic speaker prediction/segmentation
+* added first version of automatic speaker prediction for segment module
 
 Version 0.91.3
 --------------

{nkululeko-0.92.2.dist-info → nkululeko-0.93.0.dist-info}/RECORD

@@ -2,13 +2,13 @@ nkululeko/__init__.py,sha256=62f8HiEzJ8rG2QlTFJXUCMpvuH3fKI33DoJSj33mscc,63
 nkululeko/aug_train.py,sha256=FoMbBrfyOZd4QAw7oIHl3X6-UpsqAKWVDIolCA7qOWs,3196
 nkululeko/augment.py,sha256=3RzaxB3gRxovgJVjHXi0glprW01J7RaHhUkqotW2T3U,2955
 nkululeko/cacheddataset.py,sha256=XFpWZmbJRg0pvhnIgYf0TkclxllD-Fctu-Ol0PF_00c,969
-nkululeko/constants.py,sha256=HBBuhT6kpIHhRMiSBkU07cszGO7kO2A8qTYrN6zH9rw,39
+nkululeko/constants.py,sha256=YWEAJOJi8kWTdeRVHLVLQH66lH0czOfFjt-jbpnGmwY,39
 nkululeko/demo-ft.py,sha256=iD9Pzp9QjyAv31q1cDZ75vPez7Ve8A4Cfukv5yfZdrQ,770
 nkululeko/demo.py,sha256=4Yzhg6pCPBYPGJrP7JX2TysVosl_R1llpVDKc2P_gUA,4955
 nkululeko/demo_feats.py,sha256=BvZjeNFTlERIRlq34OHM4Z96jdDQAhB01BGQAUcX9dM,2026
 nkululeko/demo_predictor.py,sha256=lDF-xOxRdEAclOmbepAYg-BQXQdGkHfq2n74PTIoop8,4872
 nkululeko/ensemble.py,sha256=71V-rre61H3J4sh7lu-OTo4I2_g7mm_rQxwW1ARDHgY,12782
-nkululeko/experiment.py,sha256=h3DS-k6vk5juXa3HJXI7Z4vvnNspO4qj5SJ1o1Z3PIk,31860
+nkululeko/experiment.py,sha256=uU_8WR8JuUD50lgcl_K_BBQYmHMbuwAniWft8bGHuDU,31842
 nkululeko/explore.py,sha256=Y5lPPychnI-7fyP8zvwVb9P09fvprbUPOofOppuABYQ,3658
 nkululeko/export.py,sha256=U-V4acxtuL6qKt6oAsVcM5TTeWogYUJ3GU-lA6rq6d4,4336
 nkululeko/feature_extractor.py,sha256=UnspIWz3XrNhKnBBhWZkH2bHvD-sROtrQVqB1JvkUyw,4088
@@ -43,14 +43,14 @@ nkululeko/autopredict/ap_gender.py,sha256=b6oTqHKVwOnYh4YlKbuMflssS4HJqs_c1ayusa
 nkululeko/autopredict/ap_mos.py,sha256=e4hmgb0Yf1_AbC5P0CqXJIvufjhbTrqmI5goARxrY0Y,1107
 nkululeko/autopredict/ap_pesq.py,sha256=mRt3Loucaoy4vJxwfuxUt0fP88bMGvkmrLCEpKEXWp0,1140
 nkululeko/autopredict/ap_sdr.py,sha256=VQ2UkxOO3ipqYNNjFwKgEaGCk8IzLI5lX_2tZFLIvTY,1188
-nkululeko/autopredict/ap_sid.py,sha256=87LXMHzJ8jt2q9dUtPJd_nJi_XOcFoqpbva-BT4UJN0,1393
+nkululeko/autopredict/ap_sid.py,sha256=mCxf2DUOPUlDdnVwCeljFJtCXM4uum1poZQ9RrwHHM8,2641
 nkululeko/autopredict/ap_snr.py,sha256=AiTU8-7CMEeowmYkMO19lw1HCb1yTXC6KeulNf8gOqw,1110
 nkululeko/autopredict/ap_stoi.py,sha256=UEQg1ZV0meAsxgdWB8ieRs9GPXHqArmsaOyCGRwpcnA,1187
 nkululeko/autopredict/ap_valence.py,sha256=WrW4Ltqi_odW49_4QEVKkfnrcztLIVZ4cXIEHu4dBN8,1026
 nkululeko/autopredict/estimate_snr.py,sha256=1k9-XadABudnsNOeFZD_Fg0E64-GUQVS7JEp82MLQS4,4995
 nkululeko/data/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 nkululeko/data/dataset.py,sha256=Hz2IOsdcESG-P3aP7r4d1xj_gIP6fyGCYOwukoQ7SM8,29321
-nkululeko/data/dataset_csv.py,sha256=mb7FpHOmJHxpt1QYVBKveyqJN3MUt30TRfwwJw0TT5c,4816
+nkululeko/data/dataset_csv.py,sha256=p2b4eS5R2Q5zdOIc56NRRU2PTFXSRt0qrdHGafHkWKo,4830
 nkululeko/feat_extract/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 nkululeko/feat_extract/feats_agender.py,sha256=onfAQ6-xx_mFMJXEF1IX8cHBmGtGeX6weJmxbkfh1_o,3184
 nkululeko/feat_extract/feats_agender_agender.py,sha256=_YQv1qw--3uQfnyTQDCwlmPRnrhdMhgXbYK2yQtseW0,3464
@@ -71,8 +71,8 @@ nkululeko/feat_extract/feats_spectra.py,sha256=6WhFUpB0WTutg7OFMlAw9lSwVU5OBYCDc
 nkululeko/feat_extract/feats_spkrec.py,sha256=o_6bdU4lIkj64S5Kdjf1iyuo1VASeYxE4XdxV94a8gE,4732
 nkululeko/feat_extract/feats_squim.py,sha256=yJifsp9kj9iJjW_UAKr3LlvVhX5rv7el4bepn0wN2a8,4578
 nkululeko/feat_extract/feats_trill.py,sha256=TUCrh5xbfnHD2gzb9mlkMSV4aK6YXazMqsh5xJ5yzUI,3188
-nkululeko/feat_extract/feats_wav2vec2.py,sha256=lINWb2rBLXuMzNKV8gKsTke8wuXIF1X4jOu-GMB3aPg,5272
-nkululeko/feat_extract/feats_wavlm.py,sha256=kTuxFnymBMYP3t9yAQJjRQ5ul4AiS0O8NXq3z6B9AYs,4731
+nkululeko/feat_extract/feats_wav2vec2.py,sha256=WYB9XlRzgDi8cGSKzhV5jahA0GZ_SiWgaQ25IcEemto,5296
+nkululeko/feat_extract/feats_wavlm.py,sha256=O9cfc39VF5aPJRRATKb37pHT4W11i2cu5O1mY9LOjIA,4755
 nkululeko/feat_extract/feats_whisper.py,sha256=n3ESZtva7wshs8E8diBlQYa9xCH_P0UY1DncSrxz-FY,4508
 nkululeko/feat_extract/featureset.py,sha256=clcBv9rzBRW-bfw7JC_FYTjU5uUS-c0UE1XtQLYYRiE,1615
 nkululeko/feat_extract/feinberg_praat.py,sha256=bgzWtQkKbgcygrzwAxDXosui1rcc38qhWuJq9GLr0z8,21308
@@ -109,12 +109,12 @@ nkululeko/segmenting/seg_inaspeechsegmenter.py,sha256=b3t0zdpJYofKWMyKRMtMMX91xe
 nkululeko/segmenting/seg_pyannote.py,sha256=6IPbgjnGOz9juzEKDTZN3PSipX4t6Mz-DILAx3rp5do,4216
 nkululeko/segmenting/seg_silero.py,sha256=ulodnvtRq5MLHDxy_RmAK4tJg6h1d-mPq-uCPFkGVKg,4258
 nkululeko/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-nkululeko/utils/files.py,sha256=UiGAtZRWYjHSvlmPaTMtzyNNGE6qaLaxQkybctS7iRM,4021
+nkululeko/utils/files.py,sha256=SrrYaU7AB80MZHiV1jcB0h_zigvYLYgSVNTXV4ao38g,4593
 nkululeko/utils/stats.py,sha256=vCRzhCR0Gx5SiJyAGbj1TIto8ocGz58CM5Pr3LltagA,2948
 nkululeko/utils/util.py,sha256=XFZdhCc_LM4EmoZ5tKKaBCQLXclcNmvHwhfT_CXB98c,16723
-nkululeko-0.92.2.dist-info/LICENSE,sha256=0zGP5B_W35yAcGfHPS18Q2B8UhvLRY3dQq1MhpsJU_U,1076
-nkululeko-0.92.2.dist-info/METADATA,sha256=pwdxFGECc-W2WdmnXxgJz6Jy3CbvwzeHASfu7WxFK7g,41832
-nkululeko-0.92.2.dist-info/WHEEL,sha256=P9jw-gEje8ByB7_hXoICnHtVCrEwMQh-630tKvQWehc,91
-nkululeko-0.92.2.dist-info/entry_points.txt,sha256=lNTkFEdh6Kjo5o95ZAWf_0Lq-4ztGoAoMVSDuPtuyS0,442
-nkululeko-0.92.2.dist-info/top_level.txt,sha256=DPFNNSHPjUeVKj44dVANAjuVGRCC3MusJ08lc2a8xFA,10
-nkululeko-0.92.2.dist-info/RECORD,,
+nkululeko-0.93.0.dist-info/LICENSE,sha256=0zGP5B_W35yAcGfHPS18Q2B8UhvLRY3dQq1MhpsJU_U,1076
+nkululeko-0.93.0.dist-info/METADATA,sha256=3q74htqBYa_dfgtZzah5SPDwjG3o2c9sfGBlJK9sfI4,41933
+nkululeko-0.93.0.dist-info/WHEEL,sha256=P9jw-gEje8ByB7_hXoICnHtVCrEwMQh-630tKvQWehc,91
+nkululeko-0.93.0.dist-info/entry_points.txt,sha256=lNTkFEdh6Kjo5o95ZAWf_0Lq-4ztGoAoMVSDuPtuyS0,442
+nkululeko-0.93.0.dist-info/top_level.txt,sha256=DPFNNSHPjUeVKj44dVANAjuVGRCC3MusJ08lc2a8xFA,10
+nkululeko-0.93.0.dist-info/RECORD,,