What's Changed
Fixes
- Bug fix in MP3 and FLAC compute length on TTSDataset by @Edresson in coqui-ai#3092
- Fix get_git_branch() in generic_utils.py by Nick Potafiy in coqui-ai#3561
- fix(api): handle missing attribute in is_multi_lingual by @eginhard in eginhard#4
- Fix TTS().list_models() by @greerviau in eginhard#5
- Fix bark model url by @ipeevski in eginhard#6
- fix(xtts_manager): name_to_id() should return dict by @eginhard in eginhard#8
- fix chinese pinyin phonemes by @JaysonAlbert in eginhard#10
- Fix synthesis with fairseq models by @akgupta1337 in eginhard#11
- build: update numpy and librosa by @eginhard in eginhard#14
- docs(tts.models.vits): clarify use of discriminator/generator by @eginhard in eginhard#19
- fix: torch.stft will soon require return_complex=True by @eginhard in eginhard#20
Refactors, dependency updates
- Remove pandas dependency by @eginhard in eginhard#3
- build: remove unused dependencies by @eginhard in eginhard#12
- build: make dependencies for server optional by @eginhard in eginhard#15
- Remove duplicate Trainer code by @eginhard in eginhard#18
- refactor(dataset): get audio length with torchaudio by @eginhard in eginhard#21
- refactor(bin.find_unique_chars): use existing function by @eginhard in eginhard#22
Dev improvements
- Fix CI (save best model after 0 steps in tests) by @eginhard in coqui-ai#3423
- CI: remove unused check_skip steps by @akx in eginhard#2
- Lint overhaul (pylint to ruff) by @akx in eginhard#1
- Pin black for consistent outputs by @eginhard in eginhard#7
- test(vocoder): disable wavegrad training test in CI by @eginhard in eginhard#9
- ci: combine coverage reports and merge test workflows by @eginhard in eginhard#16
- Update readme by @eginhard in eginhard#17
- Update repository links, package names, release script by @eginhard in eginhard#24
Full Changelog: v0.22.0...v0.22.1