Block meeting start on incomplete model

Wait for live transcription to start before creating a meeting. Tighten Rust and Swift model readiness checks so partial CoreML bundles are not treated as ready.

Author: ComputelessComputer

src-tauri/src/speech_models.rs

@@ -298,26 +298,18 @@ pub fn mlx_runtime_is_available() -> bool {
 
 pub fn model_path_is_ready(model_id: SpeechModelId, path: &Path) -> bool {
     match model_id {
-        SpeechModelId::ParakeetStreaming | SpeechModelId::ParakeetBatch => required_files_present(
-            path,
-            &[
-                "config.json",
-                "vocab.json",
-                "encoder.mlmodelc",
-                "decoder.mlmodelc",
-                "joint.mlmodelc",
-            ],
-        ),
-        SpeechModelId::Omnilingual => required_files_present(
-            path,
-            &[
-                "config.json",
-                "tokenizer.model",
-                "omnilingual-ctc-300m-int8.mlpackage",
-            ],
-        ),
+        SpeechModelId::ParakeetStreaming | SpeechModelId::ParakeetBatch => {
+            required_regular_files_present(path, &["config.json", "vocab.json"])
+                && compiled_coreml_model_ready(&path.join("encoder.mlmodelc"))
+                && compiled_coreml_model_ready(&path.join("decoder.mlmodelc"))
+                && compiled_coreml_model_ready(&path.join("joint.mlmodelc"))
+        }
+        SpeechModelId::Omnilingual => {
+            required_regular_files_present(path, &["config.json", "tokenizer.model"])
+                && package_directory_ready(&path.join("omnilingual-ctc-300m-int8.mlpackage"))
+        }
         SpeechModelId::Qwen3Small | SpeechModelId::Qwen3Large => {
-            required_files_present(path, &["vocab.json", "merges.txt", "tokenizer_config.json"])
+            required_regular_files_present(path, &["vocab.json", "merges.txt", "tokenizer_config.json"])
                 && directory_contains_extension(path, "safetensors")
         }
     }
@@ -344,10 +336,35 @@ fn read_sysctl_value(name: &str) -> Option<String> {
     }
 }
 
-fn required_files_present(path: &Path, relative_paths: &[&str]) -> bool {
+fn required_regular_files_present(path: &Path, relative_paths: &[&str]) -> bool {
     relative_paths
         .iter()
-        .all(|relative_path| path.join(relative_path).exists())
+        .all(|relative_path| path.join(relative_path).is_file())
+}
+
+fn compiled_coreml_model_ready(path: &Path) -> bool {
+    path.is_dir()
+        && path.join("model.mil").is_file()
+        && directory_contains_regular_file(&path.join("weights"))
+}
+
+fn package_directory_ready(path: &Path) -> bool {
+    path.is_dir() && directory_contains_regular_file(path)
+}
+
+fn directory_contains_regular_file(path: &Path) -> bool {
+    let Ok(entries) = std::fs::read_dir(path) else {
+        return false;
+    };
+
+    entries.flatten().any(|entry| {
+        let entry_path = entry.path();
+        match entry.file_type() {
+            Ok(file_type) if file_type.is_file() => true,
+            Ok(file_type) if file_type.is_dir() => directory_contains_regular_file(&entry_path),
+            _ => false,
+        }
+    })
 }
 
 fn directory_contains_extension(path: &Path, extension: &str) -> bool {

src-tauri/swift-permissions/src/speech_bridge.swift

@@ -69,31 +69,6 @@ private enum SpeechModelKind: String, CaseIterable {
     self == .parakeetStreaming
   }
 
-  var requiredRelativePaths: [String] {
-    switch self {
-    case .parakeetStreaming, .parakeetBatch:
-      return [
-        "config.json",
-        "vocab.json",
-        "encoder.mlmodelc",
-        "decoder.mlmodelc",
-        "joint.mlmodelc",
-      ]
-    case .omnilingual:
-      return [
-        "config.json",
-        "tokenizer.model",
-        "omnilingual-ctc-300m-int8.mlpackage",
-      ]
-    case .qwen3Small, .qwen3Large:
-      return [
-        "vocab.json",
-        "merges.txt",
-        "tokenizer_config.json",
-      ]
-    }
-  }
-
   func cacheDirectoryURL() throws -> URL {
     try HuggingFaceDownloader.getCacheDirectory(for: repo)
   }
@@ -107,26 +82,30 @@ private enum SpeechModelKind: String, CaseIterable {
       return false
     }
 
-    let fileManager = FileManager.default
-    for relativePath in requiredRelativePaths {
-      if !fileManager.fileExists(atPath: directory.appendingPathComponent(relativePath).path) {
-        return false
-      }
-    }
-
-    if self == .qwen3Small || self == .qwen3Large {
-      guard let contents = try? fileManager.contentsOfDirectory(at: directory, includingPropertiesForKeys: nil)
-      else {
-        return false
-      }
-
-      return contents.contains { $0.pathExtension == "safetensors" }
+    switch self {
+    case .parakeetStreaming, .parakeetBatch:
+      return Self.regularFileExists(at: directory.appendingPathComponent("config.json"))
+        && Self.regularFileExists(at: directory.appendingPathComponent("vocab.json"))
+        && Self.compiledCoreMLModelReady(at: directory.appendingPathComponent("encoder.mlmodelc"))
+        && Self.compiledCoreMLModelReady(at: directory.appendingPathComponent("decoder.mlmodelc"))
+        && Self.compiledCoreMLModelReady(at: directory.appendingPathComponent("joint.mlmodelc"))
+    case .omnilingual:
+      return Self.regularFileExists(at: directory.appendingPathComponent("config.json"))
+        && Self.regularFileExists(at: directory.appendingPathComponent("tokenizer.model"))
+        && Self.directoryContainsRegularFile(
+          at: directory.appendingPathComponent("omnilingual-ctc-300m-int8.mlpackage")
+        )
+    case .qwen3Small, .qwen3Large:
+      return Self.regularFileExists(at: directory.appendingPathComponent("vocab.json"))
+        && Self.regularFileExists(at: directory.appendingPathComponent("merges.txt"))
+        && Self.regularFileExists(at: directory.appendingPathComponent("tokenizer_config.json"))
+        && Self.directoryContainsFile(withExtension: "safetensors", in: directory)
     }
-
-    return true
   }
 
   func load(progressHandler: ((Double, String) -> Void)?) async throws -> LoadedSpeechModel {
+    let offlineMode = filesReady()
+
     switch self {
     case .parakeetStreaming:
       return .streaming(
@@ -139,25 +118,94 @@ private enum SpeechModelKind: String, CaseIterable {
       return .parakeetBatch(
         try await ParakeetASRModel.fromPretrained(
           modelId: repo,
+          offlineMode: offlineMode,
           progressHandler: progressHandler
         )
       )
     case .omnilingual:
       return .omnilingual(
         try await OmnilingualASRModel.fromPretrained(
           modelId: repo,
+          offlineMode: offlineMode,
           progressHandler: progressHandler
         )
       )
     case .qwen3Small, .qwen3Large:
       return .qwen3(
         try await Qwen3ASRModel.fromPretrained(
           modelId: repo,
+          offlineMode: offlineMode,
           progressHandler: progressHandler
         )
       )
     }
   }
+
+  private static func regularFileExists(at url: URL) -> Bool {
+    var isDirectory = ObjCBool(false)
+    return FileManager.default.fileExists(atPath: url.path, isDirectory: &isDirectory)
+      && !isDirectory.boolValue
+  }
+
+  private static func compiledCoreMLModelReady(at directory: URL) -> Bool {
+    var isDirectory = ObjCBool(false)
+    guard FileManager.default.fileExists(atPath: directory.path, isDirectory: &isDirectory),
+      isDirectory.boolValue
+    else {
+      return false
+    }
+
+    return regularFileExists(at: directory.appendingPathComponent("model.mil"))
+      && directoryContainsRegularFile(at: directory.appendingPathComponent("weights"))
+  }
+
+  private static func directoryContainsFile(withExtension pathExtension: String, in directory: URL)
+    -> Bool
+  {
+    guard
+      let contents = try? FileManager.default.contentsOfDirectory(
+        at: directory,
+        includingPropertiesForKeys: [.isRegularFileKey]
+      )
+    else {
+      return false
+    }
+
+    return contents.contains { candidate in
+      guard
+        candidate.pathExtension == pathExtension,
+        let values = try? candidate.resourceValues(forKeys: [.isRegularFileKey])
+      else {
+        return false
+      }
+
+      return values.isRegularFile == true
+    }
+  }
+
+  private static func directoryContainsRegularFile(at directory: URL) -> Bool {
+    guard
+      let enumerator = FileManager.default.enumerator(
+        at: directory,
+        includingPropertiesForKeys: [.isRegularFileKey],
+        options: [.skipsHiddenFiles]
+      )
+    else {
+      return false
+    }
+
+    for case let candidate as URL in enumerator {
+      guard let values = try? candidate.resourceValues(forKeys: [.isRegularFileKey]) else {
+        continue
+      }
+
+      if values.isRegularFile == true {
+        return true
+      }
+    }
+
+    return false
+  }
 }
 
 private enum LoadedSpeechModel {

src/store.ts

@@ -1506,37 +1506,6 @@ function createMeeting(meetingId = crypto.randomUUID(), audioPath = "") {
   return meeting;
 }
 
-function discardMeeting(meetingId: string) {
-  const deletedActiveMeeting = currentMeetingIdFromHash() === meetingId;
-  const clearedRecordingMeeting = state.recordingMeetingId === meetingId;
-
-  clearMeetingMarkdownSync(meetingId);
-
-  state = {
-    ...state,
-    meetings: state.meetings.filter((candidate) => candidate.id !== meetingId),
-    recordingMeetingId: clearedRecordingMeeting ? null : state.recordingMeetingId,
-    transcriptionRunning: clearedRecordingMeeting ? false : state.transcriptionRunning,
-    transcriptionStopping: clearedRecordingMeeting ? false : state.transcriptionStopping,
-    liveTranscriptText: clearedRecordingMeeting ? "" : state.liveTranscriptText,
-    liveTranscriptEntries: clearedRecordingMeeting ? [] : state.liveTranscriptEntries,
-    liveTranscriptionMode: clearedRecordingMeeting ? null : state.liveTranscriptionMode,
-    meetingNote: deletedActiveMeeting ? "" : state.meetingNote,
-    diarizationMeetingId:
-      state.diarizationMeetingId === meetingId ? null : state.diarizationMeetingId,
-    diarizationIndicatorMinimized:
-      state.diarizationMeetingId === meetingId ? false : state.diarizationIndicatorMinimized,
-    diarizationBannerMessage:
-      state.diarizationMeetingId === meetingId ? null : state.diarizationBannerMessage,
-  };
-  persistMeetings();
-  emit();
-
-  if (deletedActiveMeeting) {
-    setHashRoute("/");
-  }
-}
-
 function queueMeetingAutoDiarization(meetingId: string) {
   const meeting = getMeeting(meetingId);
   if (!meeting || !meeting.audioPath.trim() || !state.diarizationSettings?.enabled) {
@@ -1813,6 +1782,10 @@ async function refreshManagedModelDownloadState(silent = false) {
   }
 }
 
+async function refreshTranscriptionSetupState() {
+  await Promise.all([refreshManagedModelDownloadState(true), refreshModelSettings(true)]);
+}
+
 async function refreshDiarizationSettings(silent = false) {
   try {
     const diarizationSettings = await invoke<DiarizationSettings>("diarization_settings_state");
@@ -1863,7 +1836,7 @@ async function refreshSettingsWindowData(silent = false) {
 }
 
 async function ensureModelReady() {
-  await refreshModelSettings(true);
+  await refreshTranscriptionSetupState();
 
   if (!state.modelSettings) {
     throw new Error("Model settings are still loading.");
@@ -2285,8 +2258,6 @@ async function startMeeting() {
     return null;
   }
 
-  let finishStartingInBackground = false;
-
   patch({
     startMeetingBusy: true,
     permissionNote: "",
@@ -2300,50 +2271,20 @@ async function startMeeting() {
     await prepareMeetingPermissions();
     await stopActiveRecordingIfNeeded();
     patch({ transcriptionBusy: true });
-    const meeting = createMeeting();
-
-    void (async () => {
-      try {
-        const snapshot = await startLiveTranscriptionSession(meeting.id);
-        const nextAudioPath = snapshot.audioPath.trim();
-        updateMeeting(meeting.id, (current) => {
-          if (current.audioPath === nextAudioPath) {
-            return current;
-          }
-
-          return {
-            ...current,
-            audioPath: nextAudioPath,
-            updatedAt: current.updatedAt,
-          };
-        });
-      } catch (error) {
-        discardMeeting(meeting.id);
-        patch({
-          permissionNote: error instanceof Error ? error.message : String(error),
-        });
-      } finally {
-        patch({
-          transcriptionBusy: false,
-          startMeetingBusy: false,
-        });
-      }
-    })();
-
-    finishStartingInBackground = true;
-    return meeting;
+    const meetingId = crypto.randomUUID();
+    const snapshot = await startLiveTranscriptionSession(meetingId);
+    return createMeeting(meetingId, snapshot.audioPath.trim());
   } catch (error) {
+    await refreshTranscriptionSetupState();
     patch({
       permissionNote: error instanceof Error ? error.message : String(error),
     });
     return null;
   } finally {
-    if (!finishStartingInBackground) {
-      patch({
-        transcriptionBusy: false,
-        startMeetingBusy: false,
-      });
-    }
+    patch({
+      transcriptionBusy: false,
+      startMeetingBusy: false,
+    });
   }
 }
 
@@ -2389,6 +2330,7 @@ async function toggleMeetingStatus(meetingId: string) {
       updatedAt: new Date().toISOString(),
     }));
   } catch (error) {
+    await refreshTranscriptionSetupState();
     patch({
       meetingNote: error instanceof Error ? error.message : String(error),
     });