CITATION.cff

cff-version: 1.2.0
message: "If you use this software or its released checkpoints, please cite it."
title: "wavepainter: Multimodal LLM-Guided Diffusion for Text-Based Speech Editing"
version: 1.0.0
date-released: 2026-08-07
license: MIT
repository-code: "https://github.com/pujariaditya/wavepainter"
abstract: >-
  A masked-span speech editor that improves on the previously reported state of
  the art in substitution and insertion word error rate, and in speaker
  similarity across substitution, insertion and deletion, on the English full
  split of Ming-Freeform-Audio-Edit. The contribution is a second training
  phase: a frozen HuBERT encoder supplies a perceptual feature loss over a
  matched crop centred on the edit, applied to a denoiser-only trainable
  surface. The released model is that phase as trained.
keywords:
  - speech editing
  - speech synthesis
  - diffusion models
  - self-supervised learning
authors:
  - family-names: "Pujari"
    given-names: "Aditya"
references:
  - type: article
    title: >-
      Edit Content, Preserve Acoustics: Imperceptible Text-Based Speech Editing
      via Self-Consistency Rewards
    year: 2026
    identifiers:
      - type: other
        value: "arXiv:2602.00560"
    notes: "The baseline these results are compared against."
    authors:
      - name: "Ren et al."