SpeechForge

0.2.x

0.2.2

11 September 2026 · Download (1.2 MB) · Release on GitHub

Added

• Authoring Language Code (default en). A bank carries no language of its own, and leaving a dub's source language to detection — which guesses from a few seconds of audio — dubbed fluent nonsense whenever it guessed wrong. Every dub now states its source language explicitly.
• Two seams a localisation add-on can fill: a bank can describe where each line's audio came from, and can produce a whole bank its own way. The Speech Library's Origin column and its Generate All button defer to them, so a specialised bank behaves correctly in the panel with no change here.

Fixed

• Speech bank queries now search subclasses. Without that, a specialised bank was invisible to the localisation status and to the line-home index — present in the project, absent from the UI.

0.2.1

8 September 2026 · Download (1.2 MB) · Release on GitHub

• Packaging fix: the release now carries everything the register allows. BuildPlugin's filter excludes Config/ and every public_extra path, so earlier zips shipped without them.
• Config/ForgeMachine.json reaches an installed copy for the first time, so the hub's Keys and Runners pages are no longer empty for it.

0.2.0

7 September 2026 · Download (1.2 MB) · Release on GitHub

• Extracted ElevenLabs and Uthana out of the core plugins, so the cores name no vendor
• Speakers, voice profiles, and the three-page Speech Library shipped, then grew to five pages (Ingest, Cast, Write, Perform, Localize)
• Casting made human-readable and controllable: audition, discovered actions, a stage
• Sessions and takes: recording stops deciding, the director chooses via a take ledger
• ConvertSpeechLine — re-voice a line without rewriting it (speech-to-speech)
• Cost estimation extended to the conversion pipeline ("price the pipeline's other meter")
• Subtitle-worthiness ("words-unverified") asked of every line, not just tool-fixable ones
• Fixed: pressing Use gave a different voice than expected, and billed for it
• Fixed: solving a face bank could overwrite a merged face
• Face bank now links to its speech bank by id instead of guessing
• Tools menu consolidated into one Automation Forge section
• Every Forge asset now has a factory; double-click opens the panel that owns it
• Localisation shipped: a sibling bank per language, joined to everything by line id
• Dubbing shipped: a dub records which recording it was made from
• All plugin links repointed to kovati.dev

0.1.x

0.1.1

29 August 2026 · Download (420 KB) · Release on GitHub

• Relicensed Apache-2.0; a release now publishes its source
• Keys moved to Editor Preferences and consolidated under ForgeKeys, then again onto the hub
• Every plugin descriptor made to agree with its release tag and credit its author
• Shared-plugin dependency made optional at build time, not just runtime

0.1.0

28 August 2026 · Download (407 KB) · Release on GitHub

• Initial release: text to a sound asset, with timings
• Runpod provisioning fix after an API change; settings relocated
• Output roots reworked so generated assets move independently of their sources
• Curated editor palette, organised by pipeline step
• Long-running solve wait proven against a multi-minute job
• Cost-before-commit pricing, and single-line banks supported