v0.1.1

Full Changelog: v0.1.0...v0.1.1

v0.1.0

What's Changed

• [ICASSP 2023] Dev update by @r9y9 in #147
• Add recipes for reproducing experiments reported in our NNSVS paper by @r9y9 in #151
• Remove recipe for no longer available data by @r9y9 in #152
• Add train_acoustic entry point in setup.py by @r9y9 in #155
• Remove vib_model and vib_flag_model from multi-stream models by @r9y9 in #156
• Fix the official URI of natsume_singing by @taroushirani in #158
• Fix config tests to cover untested acoustic model configs by @r9y9 in #159
• Replace duplicated use_harvest option by f0_extractor by @taroushirani in #160
• Fix train/dev/test split for Namine Ritsu's database by @r9y9 in #162
• Add recipe-level support for hn-uSFGAN and SiFi-GAN by @r9y9 in #157
• Add support for mcep-based aperiodicity parametrization by @r9y9 in #163
• Add guide for training DiffSinger with NNSVS's data created by stage 0 by @r9y9 in #164
• Implement has_residual_lf0_prediction method to the base class by @r9y9 in #165
• Fix oniku_kurumi_utagoe_db to use run_common_steps_dev.sh by @taroushirani in #166
• Remove hardcoded num_mels=80 and support arbitrary values by @r9y9 in #168
• Performance improvements for training acoustic models by @r9y9 in #170
• Fix MDN-based multi-stream models by @r9y9 in #172
• Expose initial value parameter for autoregressive decoder impl by @r9y9 in #173
• Deprecate unused option and add one more option to multi-stream models by @r9y9 in #174
• Support diffusion-based acoustic models by @r9y9 in #175
• preprocess: Add workaround for out-liers by @r9y9 in #176
• Add recipes for the Opencpop corpus by @r9y9 in #177
• Updated enunu2nnsvs.py to add main() by @oatsu-gh in #182
• Internal refactoring for svs.py and gen.py to make NNSVS more extensible by @r9y9 in #186
• Change not to use tempfile to avoid PermissionError in WIndows by @taroushirani in #190
• Fix CI related to matplotlib by @r9y9 in #192
• Add check for a short and single segment by @taroushirani in #193
• Update dependencies to support Python 3.10 and 3.11 by @r9y9 in #197

Full Changelog: v0.0.3...v0.1.0

v0.0.3 (2022-10-15)

What's Changed

• Fix wrong database name. by @taroushirani in #102
• Merge dev branch to master by @r9y9 in #109
• Fix Conv1dResNet bugs and add tests to prevent future unintentional bugs by @r9y9 in #111
• Refactor model.py: split it into two modules by @r9y9 in #113
• Add nn.Dropout for MDN by @r9y9 in #114
• Doc updates: split model into generic and acoustic ones by @r9y9 in #115
• Add recipe test for CI by @r9y9 in #116
• Revert #114 and add tests for backward compatibility by @r9y9 in #117
• Add MDNv2: MDN (v1) + dropout by @r9y9 in #118
• Implement correct_vuv_by_phone using C-VUV_Voiced and C-VUV_Unvoiced flags by @r9y9 in #119
• Refactor recipes by @r9y9 in #120
• Add smoothing feature for post-filtering by @r9y9 in #121
• Add GV post-filter by @r9y9 in #122
• Add correct_vuv option for data pre-processing by @r9y9 in #123
• Add more docs by @r9y9 in #124
• Support training non-resf0 acoustic models with train_resf0.py by @r9y9 in #125
• Remove MultistreamParametricModel by @r9y9 in #126
• Add tests for low-pass filter and add a corner case handing by @r9y9 in #127
• Fix error when no dynamic features are used by @r9y9 in #128
• Revisit autoregressive models: refactor Tacotron code by @r9y9 in #129
• Update docs: moved nnsvs repo to https://github.com/nnsvs org by @r9y9 in #130
• Update README.md with correct documentation link by @nicolalandro in #131
• Fix Taco inference for reduction factor by @r9y9 in #132
• Fix NonAttentiveDecoder to output (B, T, C) by @r9y9 in #133
• Fix vocoder training for different feature extraction settings by @r9y9 in #134
• Add option to allow filtering of long segments by @r9y9 in #135
• Fix bug of corect_vuv option for pre-processing by @r9y9 in #136
• Fixes for scipy 1.9.0 by @r9y9 in #137
• Wrong device assignment of svs.py by @taroushirani in #138
• Dev update by @r9y9 in #139
• Fix bug of diff vibrato mode by @r9y9 in #141
• Enable codecov by @r9y9 in #143
• Add codecov ignore config by @r9y9 in #144

Full Changelog: v0.0.2...v0.0.3

v0.0.2 (2022-04-29)

A version that should work with https://github.com/oatsu-gh/ENUNU/releases/tag/v0.4.0

New features

• Improved timings with MDN duration models #80
• Improved acoustic models with residual F0 prediction #76

Bug fixes

• numpy.linalg.LinAlgError in MDN models #94

v0.0.1 (2022-03-11)

NOTE: Tagging a new release for upcoming huge refactoring and new features (could be with breaking API changes).

The initial version of nnsvs (with some experimental features like vibrato modeling and data augmentation). This version should be compatible with currently available tools around nnsvs (e.g., ENUNU). Hydra >=v1.0.0, <v1.2.0 is supported.

PyPi release is also available. So you can install the core library by pip install nnsvs.