v1.18.0

NVIDIA/NeMo-Speech.cppv1.18.0Sep 8, 2026by abimaelmartell

AI Summary

This release focuses on improving PDF text geometry, superscript/subscript handling, and PDF loading capabilities. It also includes performance optimizations for CI and release processes, alongside various bug fixes for cross-reference tables and font metadata.

Key Highlights

Enhanced text geometry with support for rotated text bounds and baseline metadata.
Improved superscript and subscript handling for better text rendering.
Faster cross-platform CI and release publishing workflows.
Recovery of embedded bold-font metadata for accurate text representation.
Repair of classic cross-reference tables with short entries.

Breaking Changes

Positioned items and region APIs now use the visible page box instead of the CropBox, requiring removal of manual CropBox-origin adjustments.
Rust `TextItem` literals now require explicit fields: `baseline_shift`, `rotation`, and `advance_known`.

New Features

Bounded object-stream decompression to prevent memory issues.
Password-aware Python fixture tests.
Python binding tests integrated into the CI pipeline.
Positions and regions now calculated relative to the visible page box.

Full Release Notes

Improved text geometry, superscript/subscript handling, and PDF loading. Changes since **1.17.0**, released from [#509](https://github.com/firecrawl/pdf-inspector/pull/509).

## Included PRs

- [#452](https://github.com/firecrawl/pdf-inspector/pull/452): Faster cross-platform CI.
- [#453](https://github.com/firecrawl/pdf-inspector/pull/453): Faster release publishing.
- [#478](https://github.com/firecrawl/pdf-inspector/pull/478): Bounded object-stream decompression.
- [#488](https://github.com/firecrawl/pdf-inspector/pull/488): Superscript/subscript handling and baseline metadata.
- [#489](https://github.com/firecrawl/pdf-inspector/pull/489): Password-aware Python fixture tests.
- [#490](https://github.com/firecrawl/pdf-inspector/pull/490): Python binding tests in CI.
- [#487](https://github.com/firecrawl/pdf-inspector/pull/487): Positions and regions relative to the visible page box.
- [#486](https://github.com/firecrawl/pdf-inspector/pull/486): Rotated-text bounds and rotation metadata.
- [#507](https://github.com/firecrawl/pdf-inspector/pull/507): Embedded bold-font metadata recovery.
- [#508](https://github.com/firecrawl/pdf-inspector/pull/508): Repair classic cross-reference tables with short entries.

## Migration

- Positioned items and region APIs now use the visible page box. Remove any manual CropBox-origin adjustment; positioned items use a lower-left origin, while region inputs use a top-left origin.
- Rust `TextItem` literals must add `baseline_shift`, `rotation`, and `advance_known` (`0.0`, `0.0`, and `true` for ordinary upright text).

[Full changelog](https://github.com/firecrawl/pdf-inspector/blob/v1.18.0/CHANGELOG.md) · [Compare with 1.17.0](https://github.com/firecrawl/pdf-inspector/compare/32555d23356f7892762a38edb3a5b0cad7ec326b...v1.18.0)