FluidAudio currently points its built-in Silero VAD model name and manual-loading documentation at:
silero-vad-unified-256ms-v6.0.0.mlmodelc
SaySee has been testing the newer 256 ms unified Silero VAD Core ML artifact built from Silero v6.2.1 with FluidAudio's VadManager(config:vadModel:) integration. It is a drop-in artifact naming update for FluidAudio once the model bundle is available in the Hugging Face model repo.
I opened a Hugging Face model repo PR with the compiled .mlmodelc bundle here:
I also opened a FluidAudio code PR that updates:
ModelNames.VAD.sileroVadDocumentation/VAD/GettingStarted.md
Important sequencing note: the FluidAudio code PR should land after, or together with, the Hugging Face model PR. If the code change lands before the model artifact is present on FluidInference/silero-vad-coreml, the default model download path will look for a bundle that is not on the main branch yet.
SaySee-side artifact checksums for reference:
coremldata.bin: 7db35a4fd995222a7fb0129713473b15d1462572ab4a2e5e4d56bcaad9e40f41weights/weight.bin: 53ecc8b5081146140ab654c89109cf001f2183abddd7a2411c5081feeffff063model.mil: c6a9d1bf22d413265da0a07a1d14151c3ea2fad296b3aa5859275b33ef1c3270
Local verification on the FluidAudio code PR branch:
git diff --checkswift build --target FluidAudio
One broader test command currently does not complete because the package build hits an unrelated existing CLI type-checking issue in Sources/FluidAudioCLI/Commands/ASR/Parakeet/Streaming/NemotronMultilingualFleursBenchmark.swift:790 when running swift test --filter VadSegmentationTests.
FluidAudio currently points its built-in Silero VAD model name and manual-loading documentation at:
silero-vad-unified-256ms-v6.0.0.mlmodelcSaySee has been testing the newer 256 ms unified Silero VAD Core ML artifact built from Silero v6.2.1 with FluidAudio's
VadManager(config:vadModel:)integration. It is a drop-in artifact naming update for FluidAudio once the model bundle is available in the Hugging Face model repo.I opened a Hugging Face model repo PR with the compiled
.mlmodelcbundle here:I also opened a FluidAudio code PR that updates:
ModelNames.VAD.sileroVadDocumentation/VAD/GettingStarted.mdImportant sequencing note: the FluidAudio code PR should land after, or together with, the Hugging Face model PR. If the code change lands before the model artifact is present on
FluidInference/silero-vad-coreml, the default model download path will look for a bundle that is not on the main branch yet.SaySee-side artifact checksums for reference:
coremldata.bin:7db35a4fd995222a7fb0129713473b15d1462572ab4a2e5e4d56bcaad9e40f41weights/weight.bin:53ecc8b5081146140ab654c89109cf001f2183abddd7a2411c5081feeffff063model.mil:c6a9d1bf22d413265da0a07a1d14151c3ea2fad296b3aa5859275b33ef1c3270Local verification on the FluidAudio code PR branch:
git diff --checkswift build --target FluidAudioOne broader test command currently does not complete because the package build hits an unrelated existing CLI type-checking issue in
Sources/FluidAudioCLI/Commands/ASR/Parakeet/Streaming/NemotronMultilingualFleursBenchmark.swift:790when runningswift test --filter VadSegmentationTests.