Added #

• Added plain text, JSON, SRT, and WebVTT transcription result exports.
• Added cancellable sequential PCM/WAV batch transcription that reuses one loaded engine and can optionally capture per-item failures.
• Added a curated model catalog with immutable upstream revisions, SHA-256 verification, and catalog-aware model downloads.

Changed #

• Hardened managed model paths, made HTTPS the download default, and added explicit model-manager client cleanup.

Added #

• Added iOS Swift Package Manager support alongside CocoaPods.
• Added API documentation for the package's exported Dart types and members.

Changed #

• Reorganized the bundled whisper.cpp core and Flutter bridge into shared native sources used by Android, CocoaPods, and Swift Package Manager.
• Reformatted Dart sources for improved readability and maintainability without intended behavior changes.

Added #

• Added reusable Responsive, Balanced, and Efficient offline transcription modes, plus an example mode selector.
• Expanded the manual benchmark to compare all three modes with rotating run order and complete schema-v3 JSON configuration reporting.
• Added benchmark WAV playback for manually cross-checking generated transcripts against the bundled source audio.
• Added a reproducible physical-device benchmark with a fixed tiny English model configuration, JFK audio fixture, multi-run statistics, transcript accuracy checks, cancellation, and copyable JSON results in the example app.

Added #

• Added example-app screenshots to the package README, showcasing offline speech-to-text, live transcription, Silero VAD, and local diarization.

Added #

• Added a dedicated local diarization example using the TinyDiarize-compatible ggml-small.en-tdrz.bin model.
• Added model download, microphone recording, progress, cancellation, and speaker-turn visualization to the diarization example.
• Added the gdx pub.dev topic for packages maintained by Gurwinder DevX.

Changed #

• Documented the experimental two-speaker TinyDiarize workflow and its speaker-turn boundary limitations.

Changed #

• Expanded the example application to download and cache the Silero VAD model.
• Enabled integrated VAD for recorded and live transcription in the example, with an enabled-by-default switch for comparing behavior without VAD.
• Documented the example's VAD setup, model requirements, and silence-filtering behavior.

Initial public release for Android and iOS.

Added #

• Offline whisper.cpp v1.9.1 transcription and translation.
• Automatic and explicit language selection.
• Greedy and beam-search decoding with configurable prompts, suppression, and thresholds.
• Segment, word, and token timestamps, token probabilities, and speaker-turn markers.
• Non-blocking inference with progress streams and cancellation.
• Microphone recording as mono 16 kHz Float32 PCM.
• Complete-recording and windowed live transcription workflows.
• Streaming transcription for arbitrary mono PCM sources with continuous resampling, overlap removal, and timestamp rebasing.
• WAV decoding with channel mixing and sample-rate conversion.
• Integrated and standalone Silero voice activity detection, including continuous VAD.
• Optional model management with resumable downloads, SHA-256 verification, discovery, listing, and deletion.
• Android support for ARM64 and ARMv7 with CPU acceleration.
• iOS Accelerate and Metal support with optional Core ML encoder acceleration.
• Example application demonstrating model download, recording transcription, live transcription, progress reporting, and cancellation.