Skip to content

v0.4.0

Choose a tag to compare

@github-actions github-actions released this 28 Apr 15:37
· 73 commits to main since this release

✨ New Features

  • edad6c8 - enhance demo command with port, share, and server name options (commit by @deltork)
  • 81f2344 - implement text splitting (of wavs) (commit by @jordanxlau)
  • a66a443 - optionalize text splitting (commit by @jordanxlau)
  • e98c82f - wizard: echo the chosen value from menu prompts to leave a visible trace (commit by @joanise)
  • 0651468 - wizard: decent draft of question for custom g2p engines (commit by @joanise)
  • cb6c684 - wizard: allow the user to specify a custom language code (commit by @joanise)
  • 01722cf - cli rename speaker in checkpoint (commit by @deltork)
  • 319a71b - move checkpoint related cli operations to new sub-group checkpoint (commit by @deltork)
  • 9dfd8bf - util function to test if char is sentence final (commit by @jordanxlau)
  • 42ff715 - define boundaries in TextConfig (commit by @jordanxlau)
  • 0a59f84 - char and phone length stats calculated (commit by @jordanxlau)
  • a197fcf - config file for everyvoice demo (PR #729 by @deltork)
  • e81f61a - remind user when input wav has more than 2 channels (PR #707 by @MENGZHEGENG)
  • cabc602 - support dataset-specific cleaners and to_replace in TextConfig (commit by @joanise)
  • b104290 - have the wizard set dataset-specific and global cleaners (commit by @joanise)
  • 15ddb4a - apply dataset-specific cleaner and replace rules in preprocessing (commit by @joanise)
  • 436eef0 - add per language cleaners and text replacements (commit by @joanise)
  • 880de61 - have the wizard write language-specific cleaners (commit by @joanise)
  • f2f1e1f - use language-specific cleaners but remove redundant ones (commit by @joanise)
  • 11dfd99 - add detailed missing audio file reporting to preprocessor (commit by @MENGZHEGENG)
  • 4c74ff4 - let the main command groups display help when called with no args (commit by @joanise)
  • b2a56d5 - call sox as a subprocess since torchaudio is eliminating their interface (commit by @joanise)
  • 8b0208e - remove sox error on Windows since sox is now supported (commit by @joanise)
  • 9fbf410 - allow customizing the spinner text (commit by @joanise)
  • e44fd0b - improve the detection of model types in checkpoint inspect (commit by @joanise)

🐛 Bug Fixes

  • 94d6a76 - adjust regress-mix to updated configs from other suites
  • adc0538 - splice DeepForcedAligner out of EveryVoice (commit by @joanise)
  • ccf381e - update submodules to take into account duration control (commit by @joanise)
  • 07dfc02 - remove deprecated shell_complete that we no longer need anyway (commit by @joanise)
  • 7ff3557 - when installing from PyPI, let "everyvoice test" error out with a message (commit by @joanise)
  • 391e226 - let subsample.py accept -h as synonym for --help, and make it executable (commit by @joanise)
  • 258a2d7 - update submodule to accept a single line input file in synthesize (commit by @joanise)
  • 502234b - tests: give unittest discovery the right top-level directory (commit by @joanise)
  • a6c699d - suggestions based on review (commit by @jordanxlau)
  • 72b82ba - [ and ] need to be escaped for rich.print to show them (commit by @joanise)
  • e5a0c58 - tests: silence logs emitted by test_config (commit by @joanise)
  • 9e9f8ab - wizard: have input always print the prompt on stdout, never stderr (commit by @joanise)
  • 57f388d - wizard: improve custom g2p questions with MT feedback (commit by @joanise)
  • 0928d02 - test: adapt unit tests to custom g2p changes in the wizard (commit by @joanise)
  • f930b1d - improve the custom g2p prompt with MT and AP feedback (commit by @joanise)
  • 4979e66 - use questionary.text in Wizard (commit by @jordanxlau)
  • 5f72e9a - straight CLI user errors should raise typer.BadParameter (commit by @joanise)
  • dd0cc14 - tests: make cli unit tests pass again and more robust to screen width (commit by @joanise)
  • e07e51d - tests: make the malformed config test more robust (commit by @joanise)
  • 061aef4 - ci: as per CodeQL warning, declare permissions in all workflows (PR #748 by @joanise)
  • 19736a6 - regression resume files need the custom g2p step (PR #749 by @joanise)
  • 8c3690f - make make-everyvoice-env compatible with older uv (PR #757 by @joanise)
  • f5e594d - tests: relative imports are not allowed withing ev.tests (commit by @joanise)
  • 9933358 - silence some noisy tests (commit by @joanise)
  • 3323c41 - address some typing issues raised by Pylance and mypy (commit by @joanise)
  • db535fa - correct some more pyright-raised annotation errors (commit by @joanise)
  • e986ceb - pass Pydantic Field's default by name for pyright and VSCode (commit by @joanise)
  • f506c4c - fix a lot more issues detected by mypy (commit by @joanise)
  • ab07fbd - dtype=float is not recommended, use torch.float (commit by @joanise)
  • 49cfd9c - use the dataset name as label and make sure it is unique (commit by @joanise)
  • bf9808e - a few more type checking improvements (commit by @joanise)
  • 5ef0522 - remove unreachable missing files fallback code (commit by @MENGZHEGENG)
  • 6db3029 - merge multichannel and missing files tracking features (commit by @MENGZHEGENG)
  • 9e71453 - correct contact parameter in missing files tests (commit by @MENGZHEGENG)
  • b4c7e9f - apply improvements from PR review feedback (commit by @joanise)
  • ebb260c - tests: remove some cases where failing tests caused core dumps (commit by @joanise)
  • 67cf5ee - pin fastapi<0.119 to restore compatibility with Pydantic 2.7 (commit by @joanise)
  • faefa80 - ci: skip llvmlite in licensecheck since it causes a crash (PR #769 by @joanise)
  • f142949 - deps: bump nltk to version 3.9.3 (PR #774 by @marctessier)
  • 9f8525d - replace deprecated use of BaseModel.json() by current model_dump_json() (commit by @joanise)
  • c9e0ce9 - include sox_error counter in final output report (commit by @joanise)
  • addcdda - raise a meaningful exception when sox is not found (commit by @joanise)
  • 5b4c845 - subsample needs to declare encoding on open, for Windows (commit by @joanise)
  • 96d5072 - make schemas generation Windows compatible (commit by @joanise)
  • 786f4d3 - tests: make tests compatible with Windows (commit by @joanise)
  • 1d7b79a - rename TestStep to StepStub so pytest does not think it is a test (commit by @joanise)
  • 905b3b2 - use string defaults so schemas can include them, but validate them (commit by @joanise)
  • 64e4860 - tests: disable silence_c_std* when running with pytest (commit by @joanise)
  • 271d510 - deps: fix build and test dependencies (commit by @joanise)
  • d5fda0f - make-everyvoice-env should bypass hooks if already there (commit by @joanise)
  • ba3ee6d - tests: improve the pytest config and hide irrelevant warnings (commit by @joanise)
  • b900ced - deps: tensorboard 2.20.0 still needs setuptools<80 for pkg_resources (commit by @joanise)
  • 080794c - build: stop looking for libsox.so since we no longer need it (commit by @joanise)
  • a68efa2 - ci: licensecheck ignore hf-gradio, MIT but not autodetected (commit by @joanise)
  • 4b871b7 - tests: do not assume the sandbox is always called EveryVoice (commit by @joanise)
  • c91db12 - more pep585 with ruff check --select UP006,UP035 (commit by @SamuelLarkin)
  • 252563a - tests: make tests compatible with recent click versions (commit by @joanise)
  • c1eba30 - correctly find IPA lang code for sal-apa and oji-syl (PR #790 by @joanise)
  • 36bb508 - ci: use setup-ffmpeg v3.1 as v2 seems definitely broken (commit by @joanise)
  • 538eba3 - demo and app to output better error messages in case of bad models (commit by @joanise)

⚡ Performance Improvements

  • 79489e1 - wizard: defer loading g2p so that "everyvoice new-project" starts faster (commit by @joanise)
  • d60b373 - ci: cache the pip cache in the conda-macos workflow (commit by @joanise)
  • 05917b8 - ci: split the test workflow into three parallel workflows (commit by @joanise)
  • a65a321 - ci: cache apt installed packages (commit by @joanise)
  • 6cf9bc1 - tuck away imports not needed to display cli help (commit by @joanise)

♻️ Refactors

  • 50d00ac - load_config_from_path function for TextConfig (commit by @jordanxlau)
  • 7c2fb8d - implement inference base and interface (commit by @jordanxlau)
  • 1cdf4e3 - make g2p engine validation reusable (commit by @joanise)
  • 5947160 - wizard: moving text processing after selecting g2p (commit by @joanise)
  • 830d249 - tests: move all custom g2p tests to one file (commit by @joanise)
  • 9f59b4f - checkpoint related cli operations refactored to new sub-group checkpoint (commit by @deltork)
  • 5f874ce - tests: replace patch_input with patch_questionary (commit by @jordanxlau)
  • 07f653c - scaler implements append() so you don't need to use .data (PR #745 by @joanise)
  • 126e62d - tests: replace basic_test_case by the two constants it defines (commit by @joanise)
  • fc1992b - tests: flatten logs instead of assertRegex everywhere (commit by @joanise)
  • 9cd35c5 - some type checking improvements (commit by @joanise)
  • bce6e7e - use list[] instead of typing.List[], ditto dict (commit by @joanise)
  • a09b135 - use the now recommended Annotated for typer options (commit by @joanise)
  • d4ca8e0 - convert lots more options to the recommend Annotated syntax (commit by @joanise)
  • 99b29be - factor out typer file opts into typer_file_option/arg (commit by @joanise)
  • ba5f1da - use the sox CLI instead of the deprecated torchaudio wrapper (commit by @joanise)
  • 96c97d7 - tests: move sox effects tests into their own class (commit by @joanise)
  • dc17ba9 - remove all deprecated typing aliases (commit by @joanise)
  • 8bca2e5 - remove redundant list() for argument of sorted() (commit by @joanise)
  • 1f1b91c - tests: factor out generating dummy fp and vocoder models (commit by @joanise)

✅ Tests

🔧 Chores