b10326

ggml-org/llama.cppb10326Aug 7, 2026by github-actions[bot]same day as b10322

What changed since b10322

low risk

TTS timing output now includes the vocoder pass, making reported totals and audio-to-process ratios accurate.

Improvements

  • TTS timing measurement in get_output now covers the deferred waveform/vocoder work, so reported totals reflect real processing cost.
  • Audio-to-process ratio reported by the TTS pipeline is now consistent with actual elapsed time.

Bug fixes

  • Fixed understated TTS timing totals that previously omitted the vocoder pass from the reported figures.

Upgrade notes

No configuration or API changes are required; this is a reporting-only change. Expect TTS timing lines and audio-to-process ratios to look slower or larger than before, because they now include vocoder work that was previously excluded. Only the TTS timing change is documented between these builds, so other intermediate commits may exist.

Analyzed Oct 29, 2026

Upgrading?

Release notes summary

Improves the accuracy of Text-to-Speech (TTS) timing reporting by accounting for the vocoder pass in the pipeline.

Key Highlights

  • Updated timing logic in `get_output` to accurately measure the full waveform processing time.
  • Ensures reported total times and audio-to-process ratios are honest and reflective of actual processing.

Release Notes (excerpt)

<details open> tts: account for the vocoder pass in the timings line (#26733) get_output runs the waveform work the pipeline defers to it, from a single trailing window to a full pass depending on the model. Measuring it keeps the reported total and the audio to process ratio honest. </details> **Website:** - <https://llama.app> **macOS/iOS:** - macOS Apple Silicon (arm64) - macOS Apple Silicon (arm64, KleidiAI enabled) DISABLED - macOS Intel (x64) - iOS XCFramework **Linux:**…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories