idiap/coqui-ai-TTS v0.28.0

on GitHub

What's Changed

Features

• ⚠️ Massively simplify model initialization by @eginhard in #508. See that PR or updated training recipes for details.
• Download models via huggingface_hub to gracefully handle interrupted downloads and model updates by @eginhard in #544
• Add Italian text cleaner by @mraveri in #533
• Add Serbian common_voice training recipe and Vits model by @DarkoMilosevic86 in #581 and #573

Fixes

• Updates for coqui-ttts-trainer>=0.4.0 by @eginhard in #560
• Support transformers>=5.1 by @eginhard and @ROBERT-MCDOWELL in #560
• Unify model conditioning inputs by @eginhard in #537
• Improve import/startup time by @eginhard in #554
• Ensure that Pytorch is installed in Docker images by @eginhard in #554
• Fix TTS/bin scripts, small refactors by @eginhard in #529
• Fix Tortoise by @eginhard in #538
• fix(vits): multi-GPU training works again by @eginhard in #560
• fix(vits): update onnx opset version by @DarkoMilosevic86 in #565
• fix(vits): remove weight norm before onnx export by @Cassian433 in #595
• fix(vits): move YourTTS speaker encoder to correct device by @eginhard in #560
• fix(vits): convert tensors to float32 first by @eginhard in #560
• fix(bark): check if voice_dir exists before reading it by @eginhard in #560
• Reject unsupported model types in TTS API and server with clear errors by @deekshaNVIDIA in #569
• fix(xtts): preserve the final overlap in streaming inference by @kaNEGY in #591
• Fix voice cache lookup for IDs with dots/spaces (slugify on load) by @iamgrootns in #590

New Contributors

• @mraveri made their first contribution in #533
• @DarkoMilosevic86 made their first contribution in #565
• @deekshaNVIDIA made their first contribution in #569
• @kaNEGY made their first contribution in #591
• @iamgrootns made their first contribution in #590
• @Cassian433 made their first contribution in #595
• @ROBERT-MCDOWELL made their first contribution in #560

Full Changelog: v0.27.5...v0.28.0