Changelog: TypeWhisper-Nhi

All notable changes to this project will be documented in this file.

[1.2.0] - 2026-06-27

• Integrated offline Vietnamese ASR (speech-to-text) and punctuation restoration models.
• Added local offline speaker diarization (speaker splitting/labeling) using pure ONNX pyannote-community and ResNet embedders.
• Added download_models.py script to automate downloading and setting up local models (Zipformer-30M, Zipformer-68M, ViBERT punctuation, and speaker diarization models).
• Added convert_to_local_wav function in proxy.py to auto-convert incoming audio files to 16kHz mono WAV format.
• Exposed local model endpoints: local-zipformer-vi-30m, local-zipformer-vi-30m-punc, local-zipformer-vi-30m-diar, local-zipformer-vi-30m-punc-diar, local-zipformer-vi-68m, local-zipformer-vi-68m-punc, local-zipformer-vi-68m-diar, local-zipformer-vi-68m-punc-diar (supports speaker count suffixes like -spk2).
• Updated install.sh to install all local engine dependencies (onnxruntime, transformers, sentencepiece, scikit-learn, etc.).
• Automatically generate model tokens.txt from sentencepiece models.

[1.1.0] - 2026-06-25

• Renamed the proxy to TypeWhisper-Nhi for better recognition.
• Fixed the date suffix regex parsing bug (where date endings like -15 of Omni models were matched as speaker count suffixes).
• Filtered out all unused/unusable models (DeepSeek, GLM, and all -realtime WebSocket models) from the dropdown list to declutter the UI.
• Appended qwen3.5-omni-flash and qwen3-asr-flash to the models list.
• Added API key logging to api_key.txt and raw JSON logging for API responses.

[1.0.1] - 2026-06-23

• Integrated Obsidian export system.
• Added automatic VAD and chunking configuration for large audio files.
• Refined speaker diarization formatting output.

[1.0.0] - 2026-06-15

• Initial release of the local speech transcription proxy server.
• Installed WhisperKit support.