v1.3.3

What's Changed

• Improve the qwen3-asr built-in language detection logic by @gaspardpetit in #285
• Allow non-ascii in json by @gaspardpetit in #286
• Expose language detection parameters by @gaspardpetit in #287
• Centralize language tag regolution by @gaspardpetit in #288
• Allow reuse of model across multiple items by @gaspardpetit in #289
• allow reuse of qwen3 word alignment model by @gaspardpetit in #290
• Added metrics normalizer (from whisper) by @gaspardpetit in #291
• added support for non-pcm wav audio by @gaspardpetit in #292
• Fix pyright by @gaspardpetit in #293
• added support for voxtral by @gaspardpetit in #294
• Added switchlingua benchmark by @gaspardpetit in #295
• utterance acknowledgement should be whitespace tolerant by @gaspardpetit in #296
• normalize audio before senko diarization by @gaspardpetit in #297

Full Changelog: v1.3.2...v1.3.3