theo/sst
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: add ModelStorage and ModelCatalog; transcriber uses ModelStorage

Browse Source 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
This commit is contained in:
theo
2026-10-01 15:33:22 +02:00
co-authored by Claude Sonnet 5.5
parent fbc90cbd12
commit ae20008679
3 changed files with 149 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Sources/SstWhisper/WhisperKitTranscriber.swift
+13 -34
View File
@@ -10,44 +10,24 @@ public final class WhisperKitTranscriber: Transcriber {
}
private let modelName: String
private let storage: ModelStorage
private var pipeline: WhisperKit?
/// Download progress (0...1) during the one-time model download. May be called from any thread.
public var onProgress: ((Double) -> Void)?
public init(model: String = "large-v3-v20240930_turbo") {
public init(model: String = ModelCatalog.defaultVariant, storage: ModelStorage = ModelStorage()) {
self.modelName = model
self.storage = storage
}
/// Hugging Face hub root; WhisperKit stores models in `<root>/models/<repo>/<variant>`.
/// Kept in Application Support so it is neither in ~/Documents (TCC prompt, iCloud) nor re-fetched.
private static var storageBase: URL {
FileManager.default.urls(for: .applicationSupportDirectory, in: .userDomainMask)[0]
.appending(path: "Sst/models", directoryHint: .isDirectory)
}
public var variant: String { modelName }
private static let repo = "argmaxinc/whisperkit-coreml"
/// A previously downloaded variant folder, if it contains the compiled Core ML models.
private func localModelFolder(base: URL) -> URL? {
let repoDir = base.appending(path: "models/\(Self.repo)", directoryHint: .isDirectory)
let folders = (try? FileManager.default.contentsOfDirectory(atPath: repoDir.path)) ?? []
return folders
.filter { $0.hasSuffix(modelName) }
.map { repoDir.appending(path: $0, directoryHint: .isDirectory) }
.first { folder in
["MelSpectrogram", "AudioEncoder", "TextDecoder"].allSatisfy {
FileManager.default.fileExists(
atPath: folder.appending(path: "\($0).mlmodelc/coremldata.bin").path)
}
}
}
private func loadPipeline(from folder: URL, base: URL) async throws -> WhisperKit {
private func loadPipeline(from folder: URL) async throws -> WhisperKit {
try await WhisperKit(WhisperKitConfig(
downloadBase: base,
downloadBase: storage.base,
modelFolder: folder.path,
tokenizerFolder: base,
tokenizerFolder: storage.base,
prewarm: true,
load: true,
download: false
@@ -58,22 +38,21 @@ public final class WhisperKitTranscriber: Transcriber {
/// once downloaded, launching needs no network.
public func load() async throws {
guard pipeline == nil else { return }
let base = Self.storageBase
try FileManager.default.createDirectory(at: base, withIntermediateDirectories: true)
try FileManager.default.createDirectory(at: storage.base, withIntermediateDirectories: true)
if let folder = localModelFolder(base: base),
let loaded = try? await loadPipeline(from: folder, base: base) {
if let folder = storage.folder(for: modelName),
let loaded = try? await loadPipeline(from: folder) {
pipeline = loaded
return
}
let folder = try await WhisperKit.download(
variant: modelName,
downloadBase: base,
from: Self.repo
downloadBase: storage.base,
from: ModelStorage.repo
) { [onProgress] progress in
onProgress?(progress.fractionCompleted)
}
pipeline = try await loadPipeline(from: folder, base: base)
pipeline = try await loadPipeline(from: folder)
}
public func transcribe(_ samples: [Float]) async throws -> String {