mediapipe_audio 0.1.3

Google's MediaPipe Audio Classifier for Flutter on Android, iOS, macOS, Linux, Windows and the web, with Google's pinned YAMNet model.

0.1.3

• The README has an installation section using flutter pub add.
• With mediapipe_core 0.1.3, downloads of models and native libraries wait out short network failures.

0.1.2

Release 0.1.2.

0.1.1

Adds Google's NOTICE at the package root.

0.1.0

The first release.

• Audio Classifier on Android, iOS, macOS, Linux, Windows and the web, through Google's official runtimes, classifying PCM clips (AudioData, with a WAV decoder) at any sample rate.
• On 32-bit Android (armeabi-v7a) the Audio Classifier runs on the CPU.
• Audio stream mode on every platform: AudioClassifierOptions(runningMode: AudioRunningMode.audioStream) creates a task that takes blocks of any length with classifyAsync(block, timestampMilliseconds:) and delivers one result per model window on results, stamped with where the window's audio starts; dispose() classifies the tail. Google's own stream runs it on Android, iOS, macOS, Linux and Windows; in browsers, where Google has none, the package emulates it on Google's clips mode. The rate, channel and timestamp checks run in Dart with one message on every platform. Outside browsers at most 64 streams can be open at once. classify on a stream task, and classifyAsync or results on a clips task, throw StateError.
• AudioClassifierResult has classifications (one Classifications per model head) and timestampMilliseconds. AudioClassifierOptions extends core's TaskOptions with Google's classifier settings and runningMode.
• AudioModels.byName names YAMNet for hooks.user_defines.mediapipe_audio.models, which dart run mediapipe_core:bundle_models bundles into the app. model: uses the bundled copy and downloads at run time only when ModelStore.allowDownloads is true; app-supplied paths and bytes work too.