- Home
- Text-to-Speech Software
- VoiceCraft-X
On the recordRead 13 Oct 202622 facts
VoiceCraft-X
On the record
- Authorship
- The repository credits Zhisheng Zheng and the University of Texas at Austin with copyright in 2025.github.com · read 13 Oct 2026
- Availability
- The repository's listed inference examples are speech editing and speech synthesis Jupyter notebooks.github.com · read 13 Oct 2026
- Checkpoints
- The repository links to a default multilingual checkpoint and English and Chinese monolingual checkpoints, with more languages noted as forthcoming.github.com · read 13 Oct 2026
- Code-switching
- The site shows Chinese-English code-switching samples and says Spanish-English and French-English samples are coming soon.voicecraft-x.github.io · read 13 Oct 2026
- Commercial use
- Novoicecraft-x.github.io · read 20 Sep 2026
- Consent
- The repository says speech generation or editing should not be performed using someone's speech without that person's consent.github.com · read 13 Oct 2026
- Development status
- The repository lists environment setup, inference code, additional monolingual checkpoints, hosted demos, Colab notebooks, command-line support, and efficiency improvements as TODO items.github.com · read 13 Oct 2026
- Fine-tuning
- A code-switching sample table includes a VoiceCraft-X (10h fine-tuning) column.voicecraft-x.github.io · read 12 Oct 2026
- Input and training
- The system overview caption says the model takes text and a speaker embedding as input and is trained to predict speech-token sequences.voicecraft-x.github.io · read 12 Oct 2026
- Languages
- The site lists English, Mandarin, Korean, Japanese, Spanish, French, German, Dutch, Italian, Portuguese, and Polish.voicecraft-x.github.io · read 13 Oct 2026
- License
- The work is licensed under Creative Commons Attribution-NonCommercial 4.0 International, and the repository says to contact the authors for commercial use.github.com · read 13 Oct 2026
- Model architecture
- The site says VoiceCraft-X uses the Qwen3 large language model for phoneme-free cross-lingual text processing.voicecraft-x.github.io · read 12 Oct 2026
- Product
- VoiceCraft-X is an autoregressive neural codec language model for multilingual speech editing and zero-shot text-to-speech synthesis.voicecraft-x.github.io · read 12 Oct 2026
- Purpose
- VoiceCraft-X combines multilingual speech editing with zero-shot text-to-speech synthesis in one autoregressive neural codec language model.voicecraft-x.github.io · read 13 Oct 2026
- Setup
- The repository installation instructions create a Python 3.10 Conda environment and install dependencies with pip.github.com · read 13 Oct 2026
- Speech editing
- The site presents examples of editing existing speech by changing or adding words in a transcript.voicecraft-x.github.io · read 12 Oct 2026
- Speech editing dependency
- Montreal Forced Aligner is used for speech editing, and the repository says it can be skipped if installation is difficult.github.com · read 13 Oct 2026
- Speech generation
- Its token reordering mechanism uses time-aligned text and speech tokens to frame editing and synthesis as sequence generation.voicecraft-x.github.io · read 13 Oct 2026
