v0.6.5: Support more versions of Python

What's Changed

• build(deps-dev): bump black from 22.12.0 to 24.3.0 by @dependabot in #64
• fix: generate py39 wheels too by @joanise in #66
• build: build Python 3.13 wheels too by @joanise in #67

Full Changelog: v0.6.4...v0.6.5

v0.6.4: Support newer Python versions

What's Changed

• Python 3.12 support by @dhdaines in #62
• Build and upload releases from GitHub CI by @dhdaines in #63

Full Changelog: v0.6.2...v0.6.4

v0.6.2; UMD and no-WASM

What's Changed

• Add UMD module and JS-only target for older browsers by @dhdaines in #58

Full Changelog: v0.6.1...v0.6.2

v0.6.0: Large cleanups, ES6 modules

This is mainly a release to clean up a bunch of issues with the code. The big changes are the inclusion of type hints in the Python API, and...

A HUGE BREAKING CHANGE

The JavaScript API is now ES6 modules only. If you really need CommonJS perhaps some accomodation can be reached. Also, Node and Web environments now get separate entry points, so less (but sadly still some) Webpack tomfoolery is necessary.

What's Changed

• Add type hints to Python API by @dhdaines in #54
• Add and use pre-commit by @dhdaines in #55
• Format all the things by @dhdaines in #56
• Switch to ES6 modules (and only ES6 modules) by @dhdaines in #57

Full Changelog: v0.5.0...v0.6.0

v0.5.0: Bug fixes and JavaScript API rework

There was a "0.4.2" which was NPM-only, but this updates all the releases at once.

• Fix a serious bug in state alignment which caused a segfault on align failure.
• Fix serious bugs in the JavaScript API leading to memory exhaustion
• Remove superfluous use of promises/async in some parts of JavaScript API
• Add a (possibly temporary) spectogram API in JavaScript
• Removed some bogus internal APIs

What's Changed

• Increase wheel production and sdist too by @dhdaines in #32
• fix: extremely bogus state align code by @dhdaines in #34
• feat: export Endpointer interface (though it will go away) by @dhdaines in #35
• feat: stricter emcc linker options in external file by @dhdaines in #41
• Remove bogus code by @dhdaines in #42
• Remove unused setjmp/logjmp by @dhdaines in #43
• Fix horrific memory corruption in process() when called with subarrays by @dhdaines in #45
• Add a method to get spectrogram data by @dhdaines in #46
• fix: WTF JavaScript... TypedArray constructor strikes again by @dhdaines in #47
• Avoid overflowing the emscripten stack (WHOOPS) by @dhdaines in #49
• Remove remaining stackAlloc by @dhdaines in #50
• Make the API less asynchronous by @dhdaines in #51
• docs: update js api docs by @dhdaines in #52
• Major JavaScript API cleanup by @dhdaines in #53

Full Changelog: v0.4.1...v0.5.0

0.4.1: Fix Windows damage

Just a quick bugfix release for a stupid WIndows problem.

What's Changed

• Work around Windows damage, run CI on lesser platforms by @dhdaines in #31

Full Changelog: v0.4.0...v0.4.1

0.4.0: API update, subword alignment, VAD

What's Changed

• Merge new features from PocketSphinx 5 (and simplify things) by @dhdaines in #27
• docs: update javascript docs by @dhdaines in #28
• Correct pointer aliasing in SWAP_FLOAT32 by @dhdaines in #29
• Final 0.4.0 release updates by @dhdaines in #30

Full Changelog: v0.3.2...v0.4.0

0.3.2: Quick bug fix

Oops, ps_add_dict() was really quite broken.

This also fixes the TypeScript declarations so as to work properly with tsc --strict.

0.3.1: mostly bug fixes

This release fixes some small (or severe, depending on your point of view) problems:

• Big-endian platforms were broken, oops!
• It wasn't possible to mix int16 and float32 input data, not that you would really want to do that.
• Various tests didn't actually work
• It turns out -remove_noise was actually useful, so it has been brought back from the dead.

0.3.0: stdio removal

Use of standard I/O is not so much removed as made non-essential. This means that the JavaScript version no longer needs the huge and clunky Emscripten FS library, and thus is quite a bit smaller (down to 200k or so of WASM and 42k of JavaScript). We also load everything with memory-mapped I/O in C or Python, which is slightly faster.

There is a -cionly option which can make alignment a lot faster and not really less accurate.

Various issues have been fixed. The API is still not stable though.