Change log¶

v0.1.0 <2021-08-11>¶

#112: Renamed continuous_dict to numeric_dict to be more precise.
#112: Bandmat is now a part of nnmnkwii’s internal package. This is to avoid instllation failusres on python >=3.6. https://github.com/MattShannon/bandmat/issues/10
#112: Started testing using github actions (python 3.7, 3.8, 3.9)
#112: [hts.io]: nnmnkwii.io.hts.load_question_set() now keeps question names (e.g. “L-Phone_Boin”) in dictionary.
#112: [hts.io]: New functionality nnmnkwii.io.hts.write_audacity_labels()
#112: [hts.io]: New functionality nnmnkwii.io.hts.write_textgrid()
#112: Renamed np.int to int
#112: Added pyproject.yaml

#99: Fix future warning from sklearn
#101: [hts][frontend] various fixes for singing voice synthesis. Our frontend now supports MIDI number extraction. HTSLabelFile supports list and slice indexing.

#88: Tentative fix: make bandmat optional requirement as it is causing installation errors on python 3.7. See [here](https://github.com/MattShannon/bandmat/issues/10) for details.
#85: Fixed rounding error in caluculating number of frames.
#87: Fixed nnmnkwii.preprocessing.trim_zeros_frames() issue. Support passing trim argument.

Fix typo: adjast_frame_length and adjast_frame_lengths are renamed to adjust_frame_length and adjust_frame_lengths, respectively,
#63: Improved support for nnmnkwii.preprocessing.adjast_frame_length() and nnmnkwii.preprocessing.adjast_frame_lengths(). Padding for 1d array is now supported.
BUG FIX: example audio data is now included in the release tar.gz

Fix RuntimeError when HTS label file has white spaces between fields. Skip comments when reading HTS labels.

#61: Misc dataset improvements. Unified max_files=None from max_files=50 and add max_files args for VCTK data sources.
#59: Bug fix for memory re-allocations when num frames exceed padded_initial_guess
#60: FileSourceDataset: better descriptive error messages
#57: Add append method to HTSLabelFile and simplify structure. frame_shift_in_micro_sec was removed from its property.
#55: Add mu-law companding/expansion
Add support for JSUT dataset ver 1.1
#20: Support for mono phone labels and fix bug of silence_phone_indices() for non-state level alignment label files.

#38: Add parameter trajectory smoothing.
#37: Add tqdm as dependency. Dataset’s asarray now report progress if verbose > 0.
#37: Add further support for incremental mean/var computation.
#37: Add and improve normalization utilities, nnmnkwii.preprocessing.inv_scale(), nnmnkwii.preprocessing.inv_minmax_scale() and nnmnkwii.preprocessing.minmax_scale_params().
Add builtin data source for Voice Conversion Challenge (VCC) 2016 dataset.
#34: Add nnmnkwii.preprocessing.adjast_frame_length().
#34: adjast_frame_lengths now supports divisible_by parameter. ensure_even is deprecated.
#34: Rename adjast_frame_length to adjast_Frame_lengths
Add references to nnmnkwii.postfilters.merlin_post_filter().

#28: Setuptools improvements. 1) __version__ now includes git commit hash. 2) description read README.rst using pandoc.
#27: Add preemphasis / inv_preemphasis
#26: Add tests for GMM based voice conversion if swap=True
#25: fix typo in nnmnkwii/baseline/gmm.py

Add tests, achieve 75% test coverage.
#23, #22: Preprocess rewrite & module restructure.
#21: Add new function nnmnkwii.autograd.UnitVarianceMLPG that can run on CPU/GPU.