Skip to content
AIBRAINWORX / OpenSourceCoder.comLOCAL-FIRST. BUILT IN THE OPEN.
Open Source / Developer Tools / Text-to-Speech

llama.cpp b12478 Adds tts_dict Support for Kokoro G2P Template

The llama.cpp b12478 release introduces tts_dict for the Kokoro G2P chat template alongside value_float_t range checks and new model metadata array helpers.

Official project or press image for llamacpp-b12478-tts-dict-kokoro-g2p-template
Official image from github.com; identification only, no endorsement.
Event / context: 2026-11-17. Source check: October 28, 2026.

The open-source inference engine project ggml-org/llama.cpp published release b12478 on November 17, 2026, introducing tts_dict support for the Kokoro grapheme-to-phoneme (G2P) chat template under pull request #32397. Tagged by contributor ngxson at commit 8d1599f, the release focuses on speech-oriented chat template handling and lower-level metadata and type-safety adjustments.

What Changed in b12478

According to the official release notes on GitHub, the update combines template additions with internal validation and metadata API refinements:

  • Kokoro G2P template dictionary: Adds tts_dict within the chat layer to support Kokoro grapheme-to-phoneme template formatting.
  • Metadata array helpers: Replaces JSON-in-metadata handling (noted as no json-in-meta) with dedicated llama_model_meta_arr_str and length helper functions (llama_model_meta_arr_str/_n).
  • Numeric bounds validation: Adds a range check for value_float_t values.
  • Test coverage: Includes additional test cases referenced in the commit log as more testing.

Why It Matters for Open-Source Developers

Grapheme-to-phoneme conversion is a foundational preprocessing step for text-to-speech pipelines, and moving dictionary data out of serialized JSON strings inside GGUF or model metadata into structured array accessors simplifies how downstream C/C++ applications inspect model metadata. For teams integrating Kokoro-based speech workflows or custom chat templates into llama.cpp, avoiding ad-hoc JSON parsing in metadata reduces boilerplate and tightens type checking through explicit float range validation.

Scope and Limitations

The release page lists the tag and commit notes for PR #32397 without publishing comparative speed benchmarks, audio quality metrics, or platform-specific test results. Developers targeting production speech pipelines should review the upstream commit diff and verify behavior with their own target models and host environments before upgrading.

EXPLORE THE SOURCE

GitHub repositories

Automated Vertex AI (gemini-4-hydrogen) primary-source report; reviewed 2026-11-17.

Follow your curiosity.

Pass the signal.