1.1.16

takahirom/arbigent1.1.16Feb 16, 2026by SWivid

AI Summary

This release focuses on improving speech editing quality and training stability. It introduces new community model support and CLI enhancements, alongside optimizations for memory usage and embedding handling.

Key Highlights

Fixed speech editing boundary artifacts by working in mel domain
Added Latvian model to shared community models list
Added support for hf:// links on CLI
Optimized DiT text embedding with batched per-sample seq handling
Configurable WandB project/run_name/resume_id via YAML

New Features

Latvian model support
hf:// links on CLI
Torch utils checkpointing in mmdit forward loop
Configurable WandB settings via YAML

Full Release Notes

## What's Changed
* Fix speech editing boundary artifacts by working in mel domain by @acadarmeria in https://github.com/SWivid/F5-TTS/pull/1242
* Adding Latvian model to shared community models list by @raivisdejus in https://github.com/SWivid/F5-TTS/pull/1250
* Adding support for hf:// links on CLI by @raivisdejus in https://github.com/SWivid/F5-TTS/pull/1252
* change prepare_csv_wavs from relative path to absolute path and get d… by @ZhikangNiu in https://github.com/SWivid/F5-TTS/pull/1256
* Ignore padding at the end of the GT mel spectrogram when training sample by @ZhikangNiu in https://github.com/SWivid/F5-TTS/pull/1261
* Use torch.utils.checkpoint in mmdit forward loop when enabled to redu… by @ZhikangNiu in https://github.com/SWivid/F5-TTS/pull/1265
* Make wandb project/run_name/resume_id configurable via yaml by @ZhikangNiu in https://github.com/SWivid/F5-TTS/pull/1266
* Optimize DiT text embedding with batched per-sample seq handling by @QingyuLiu0521 in https://github.com/SWivid/F5-TTS/pull/1267

## New Contributors
* @acadarmeria made their first contribution in https://github.com/SWivid/F5-TTS/pull/1242
* @raivisdejus made their first contribution in https://github.com/SWivid/F5-TTS/pull/1250

**Full Changelog**: https://github.com/SWivid/F5-TTS/compare/1.1.15...1.1.16