Skip to content

Build package

To build a package using WhisperRealtime plugin, you need to configure the asset manager to include the machine learning model assets in the package.

Asset manager settings

Add an element to Project Settings > Game > Asset Manager > Primary Asset Types to Scan and set the following values.
(Or, since the default element with Index=1 is already set to PrimaryAssetLabel, you can edit this element.)

  1. Set Primary Asset Types to PrimaryAssetLabel.
  2. Set Asset Base Class to PrimaryAssetLabel.
  3. Uncheck Is Editor Only.
  4. Add an element to Specific Assets and specify WhisperModel_all.

Reduce package size

Package size can be reduced by packaging only a portion of the machine learning models used by the plugin.

  • You can package only the specified model size by specifying the labels under BySize in Plugins > WhisperRealtime > PrimaryAssetLabels instead of WhisperModel_all in the Asset manager settings "4".

  • Or, by specifying the labels under ByFunctionAndLanguage, you can package only models related to the specified function.

  • Or, you can refer to the table below and manually specify only the models you need.

Relationship between ML model assets and the functions that use them

English only models

Asset name of ML model Transcription Short-phrase Alignment Long-phrase Alignment
encoder_mask_1500_0_{SIZE}_en_opt_fp161 ✅ ✅ ✅
decoder_1_-1_16_{SIZE}_en_opt_fp162 ✅
aligner_b_n_{SIZE}_en_opt_fp16 ✅ ✅

Multilingual models

Asset name of ML model Transcription Short-phrase Alignment Long-phrase Alignment
encoder_mask_1500_0_{SIZE}_opt_fp16 ✅ ✅ ✅
decoder_1_-1_16_{SIZE}_opt_fp163 ✅
aligner_b_n_{SIZE}_opt_fp16 ✅ ✅

  1. {SIZE} is tiny, base, small, or medium ↩

  2. If the model size is Medium, decoder_1_-1_8_medium_en_opt_fp16 ↩

  3. If the model size is Medium, decoder_1_-1_8_medium_opt_fp16 ↩