Open-source roundup: VoiceStudio, OpenWhispr & Axolotl

This GitHub project roundup introduces open source software for speech, model training and agent workflows, alongside desktop utilities. It gives short feature summaries rather than installation walkthroughs or comparative benchmarks.

The speaker presents VoiceStudio as a fully local alternative to ElevenLabs for voice cloning, dubbing, transcription and audiobook creation. Its model catalog routes speech engines to CPU or GPU, and its audiobook editor accepts text, EPUB and PDF. OpenWhispr offers hotkey dictation at the cursor, with local Whisper or Nvidia Parakeet transcription or a choice of cloud providers. Both apps are described as supporting macOS, Windows and Linux.

For model development, the roundup covers Heretic's automated modification of refusal behavior and Axolotl's YAML-based fine-tuning workflows. The speaker says Axolotl supports training on one GPU through multiple nodes and can run locally with Docker or in the cloud.

Utopia is presented as a self-hosted knowledge platform whose graph records when facts were valid. It combines document retrieval with cited chat answers and, according to the speaker, can run airgapped. The roundup also distinguishes remote services: HexStellar's Python client sends optimization problems to hosted servers and contains no solver itself.