From ec6752b584ff5a33a9fae894f0f947793c960229 Mon Sep 17 00:00:00 2001 From: HaD0Yun Date: Sun, 22 Mar 2026 19:25:53 +0900 Subject: [PATCH] Document the current IPA phoneme fine-tuning status Adds a note that explains the public repository does not yet document a supported Piper-compatible IPA phoneme fine-tuning workflow. Constraint: Keep this branch scoped to issue #4 only Rejected: Folding this into a multi-issue combined PR | the requested delivery shape is one PR per issue Confidence: high Scope-risk: narrow Directive: Keep this note fact-limited to what the public repository and issue thread currently support Tested: git diff --check --cached Not-tested: Runtime feature execution for the requested capability --- docs/issues/issue-4-ipa-phoneme-finetuning.md | 24 +++++++++++++++++++ 1 file changed, 24 insertions(+) create mode 100644 docs/issues/issue-4-ipa-phoneme-finetuning.md diff --git a/docs/issues/issue-4-ipa-phoneme-finetuning.md b/docs/issues/issue-4-ipa-phoneme-finetuning.md new file mode 100644 index 0000000..625277c --- /dev/null +++ b/docs/issues/issue-4-ipa-phoneme-finetuning.md @@ -0,0 +1,24 @@ +# Issue #4 — IPA phoneme fine-tuning status + +## Summary +The current repository does not document a supported fine-tuning workflow for Piper-compatible IPA phoneme training. + +## Current repository state +- The public repository is inference-focused. +- The root README documents model loading, prompting, multilingual alignment, and generation. +- The repository does not currently expose end-to-end training or fine-tuning scripts. +- The codebase contains tokenization / alignment components, but there is no documented pathway here for swapping the training target to a Piper-compatible IPA phoneme representation. + +## What this means today +- Fine-tuning on Piper-compatible IPA phonemes is currently undocumented in this repository. +- The repository does not provide a supported recipe for preparing IPA labels, retraining the aligner/tokenizer stack, or adapting the model for that representation. + +## What would be required to support it +A real support path would likely need: +1. training or fine-tuning scripts +2. data-format guidance for phoneme-aligned supervision +3. tokenizer / aligner decisions for IPA-based inputs +4. validation examples that show the resulting generation path still works + +## Practical next step +Until training code is released, the safest statement is that IPA phoneme fine-tuning is not yet documented or supported in the public repo.