A Python/Pytorch app for easily synthesising human voices
Build with GPU support: https://mega.nz/file/4hpgFBAb#6GP3p0n-s5v9KIXmFRqxDst7BPaNhEy14JMlEe0aopY
Notes: Improve transcription error logging Update dataset info when manually labelling a clip Fix quotation problem in synthesis text
Build with GPU support: https://mega.nz/file/0woADQSS#vud7UO7Pi-wNsa6eawqqdzcqNoCNGzH50XHDlbQnd0E
Notes: Add custom vocoder training Fix data path issue to ensure all files are in the data folder
Build with GPU support: https://mega.nz/file/h8wQBDJY#j4jq3PVj4LDhMPeJJwwhod1bVuve9Sf3G6NyYzo-mZg
Notes: Fix unlabelled clip playback Add new remote training notebook Improve invalid symbols error message (thanks to @SirBitesalot) Fix symbols selection for non-English dataset creation (thanks to @SirBitesalot)
Build with GPU support: https://mega.nz/file/Q5Ak2BZL#-AQTJyHO-wh0sGC5IcCX6Hjce6WQvhT7OXaueSZ2ztg
Notes: Fixed issue with invalid clips being added for manual labelling Fixed multi-line synthesis results folder naming (thanks @Marclass) Added support for custom languages to text cleaning to improve non-English dataset quality Improved dataset validation and errors
Build with GPU support: https://mega.nz/file/FkgGASaL#q8hn70t6zn6m_a9UwwwI2RWbwz3BqzaJ6b7nfxucUbg
Notes: Remove pre-trained model buildup Add manage dataset option to view dataset info & label unlabelled clips
Build with GPU support no longer available
Notes: Fix training from checkpoint issue Fix alignment GIF bug Update navigation bar (including links to docs & discord)
Build with GPU support: https://mega.nz/file/xlgXlQSD#Ls1BN_CDDyoL5cqjzBhrh63RTQAa6fUv5DYnDY0Jowo
Notes: Generate an alignment graph in training to visualise how the model is performing against a test sentence Export timelapse of alignments to visualise improvement over time Add sentence length recommendation to synthesis
Build with GPU support no longer available
Notes: Package FFmpeg & Silero with the executable to avoid additional downloads
Build with GPU support no longer available
Notes: Improve symbol weights transfer to improve training from an existing model (thanks to @CookiePPP) Add attention scoring to training Fix dataset info generation when extending existing datasets Fix issue with checkpoint selection on the training page
Build with GPU support no longer available
Notes: Add paragraph synthesis option with automatic sentence splitting Remove clip length requirement in dataset importing