feat(local-asr): 模型状态透出目录元数据、新增中断下载清理命令与 Zipformer 目录项

LocalAsrModelStatus 补齐 display_name/family/mode/languages/size_bytes，
前端无需为基础元数据实时访问 HuggingFace；新增 local_asr_cleanup_incomplete
命令（LocalAsrApi 默认实现 + Service 实现 + 前端包装）清理 staging 目录；
sherpa 目录新增 zipformer-bilingual-zh-en-streaming 流式中英模型并补齐八语文案。
修复测试里 LocalAsrTarget::parse 的 .into() 类型推断歧义。

openless-all/app/crates/openless-core/src/domains.rs

@@ -280,6 +280,12 @@ pub trait LocalAsrApi: Send + Sync {
     fn release(&self, runtime: LocalAsrRuntime) -> BoxFuture<'static, Result<(), BackendError>>;
     fn preload(&self, runtime: LocalAsrRuntime) -> BoxFuture<'static, Result<(), BackendError>>;
     fn delete_model(&self, target: LocalAsrTarget) -> BoxFuture<'static, Result<(), BackendError>>;
+    fn cleanup_incomplete(
+        &self,
+        target: LocalAsrTarget,
+    ) -> BoxFuture<'static, Result<(), BackendError>> {
+        unsupported("local ASR incomplete download cleanup")
+    }
     fn model_dir(
         &self,
         target: LocalAsrTarget,

openless-all/app/crates/openless-core/src/local_asr_service.rs

@@ -1254,6 +1254,14 @@ impl LocalAsrApi for LocalAsrService {
         })
     }
 
+    fn cleanup_incomplete(
+        &self,
+        target: LocalAsrTarget,
+    ) -> BoxFuture<'static, Result<(), BackendError>> {
+        let store = Arc::clone(&self.model_store);
+        Box::pin(async move { store.cleanup_incomplete(&target) })
+    }
+
     fn model_dir(
         &self,
         target: LocalAsrTarget,

openless-all/app/src-tauri/src/commands/local_asr.rs

@@ -76,12 +76,19 @@ impl From<openless_core::LocalAsrStorageSettings> for LocalAsrStorageSettings {
     }
 }
 
+/// 与 Sherpa `SherpaCatalogWire` 对齐：透出展示名、家族、模式、语言与远端尺寸，
+/// 前端无需再为基础元数据实时访问 HuggingFace。
 #[derive(Debug, Serialize)]
 #[serde(rename_all = "camelCase")]
 pub struct LocalAsrModelStatus {
     pub id: String,
     pub hf_repo: String,
+    pub display_name: String,
+    pub family: String,
+    pub mode: Option<String>,
+    pub languages: Vec<String>,
     pub downloaded_bytes: u64,
+    pub size_bytes: Option<u64>,
     pub is_downloaded: bool,
 }
 
@@ -89,8 +96,13 @@ impl From<openless_core::LocalAsrModel> for LocalAsrModelStatus {
     fn from(model: openless_core::LocalAsrModel) -> Self {
         Self {
             id: model.target.model_id().to_string(),
-            hf_repo: model.repository.unwrap_or_default(),
+            hf_repo: model.repository.clone().unwrap_or_default(),
+            display_name: model.display_name,
+            family: model.family,
+            mode: model.mode,
+            languages: model.languages,
             downloaded_bytes: model.downloaded_bytes,
+            size_bytes: model.size_bytes,
             is_downloaded: model.installed,
         }
     }
@@ -355,6 +367,20 @@ pub async fn local_asr_delete_model(
         .map_err(core_error)
 }
 
+/// 清理指定模型中断下载遗留的 staging 目录，不触碰已安装模型。
+#[tauri::command]
+pub async fn local_asr_cleanup_incomplete(
+    backend: CoreState<'_>,
+    model_id: String,
+) -> Result<(), String> {
+    backend
+        .services()
+        .local_asr
+        .cleanup_incomplete(target(LocalAsrRuntime::Generic, model_id)?)
+        .await
+        .map_err(core_error)
+}
+
 #[tauri::command]
 pub async fn local_asr_model_dir(
     backend: CoreState<'_>,
@@ -489,4 +515,40 @@ mod wire_contract_tests {
             })
         );
     }
+
+    #[test]
+    fn generic_model_status_keeps_the_legacy_fields_and_adds_catalog_metadata() {
+        let core = openless_core::LocalAsrModel {
+            target: openless_core::LocalAsrTarget::parse(
+                openless_core::LocalAsrRuntime::Generic,
+                "qwen3-asr-0.6b",
+            )
+            .unwrap(),
+            display_name: "Qwen3 ASR 0.6B".into(),
+            family: "qwen3_asr".into(),
+            mode: None,
+            repository: Some("Qwen/Qwen3-ASR-0.6B".into()),
+            languages: vec!["zh".into(), "en".into()],
+            installed: false,
+            downloaded_bytes: 4096,
+            size_bytes: Some(1_234_567),
+        };
+
+        let value = serde_json::to_value(LocalAsrModelStatus::from(core)).unwrap();
+
+        assert_eq!(
+            value,
+            serde_json::json!({
+                "id": "qwen3-asr-0.6b",
+                "hfRepo": "Qwen/Qwen3-ASR-0.6B",
+                "displayName": "Qwen3 ASR 0.6B",
+                "family": "qwen3_asr",
+                "mode": null,
+                "languages": ["zh", "en"],
+                "downloadedBytes": 4096,
+                "sizeBytes": 1234567,
+                "isDownloaded": false,
+            })
+        );
+    }
 }

openless-all/app/src-tauri/src/lib.rs

@@ -351,6 +351,7 @@ macro_rules! app_invoke_handler_desktop {
             commands::local_asr_download_model,
             commands::local_asr_cancel_download,
             commands::local_asr_delete_model,
+            commands::local_asr_cleanup_incomplete,
             commands::local_asr_model_dir,
             commands::local_asr_reveal_model_dir,
             commands::local_asr_reveal_models_root,

openless-all/app/src/i18n/de.ts

@@ -2129,6 +2129,9 @@ export const de: typeof zhCN = {
     sherpaModelWhisperLargeV3: 'Whisper Large V3 (mehrsprachig)',
     sherpaModelWhisperLargeV3Desc:
       'Leistungsstärkste mehrsprachige Open-Source-Whisper-Variante: hohe Qualität, großer Download.',
+    sherpaModelZipformer: 'Zipformer-Streaming (zh/en)',
+    sherpaModelZipformerDesc:
+      'Streaming-Modell für Chinesisch und Englisch mit der geringsten Latenz – Text erscheint beim Sprechen.',
     sherpaModelQwen3: 'Qwen3-ASR 0.6B INT8',
     sherpaModelQwen3Desc:
       'Konvertiertes sherpa-onnx-Qwen3-ASR-Modell mit mehrsprachiger Erkennung und besserer Verarbeitung längerer Kontexte.',

openless-all/app/src/i18n/en.ts

@@ -2063,6 +2063,9 @@ export const en: typeof zhCN = {
     sherpaModelWhisperLargeV3: 'Whisper Large V3 (multilingual)',
     sherpaModelWhisperLargeV3Desc:
       'The best open-source multilingual Whisper tier — high quality, large download.',
+    sherpaModelZipformer: 'Zipformer Streaming (zh/en)',
+    sherpaModelZipformerDesc:
+      'Streaming Chinese-English model with the lowest latency — text appears as you speak.',
     sherpaModelQwen3: 'Qwen3-ASR 0.6B INT8',
     sherpaModelQwen3Desc:
       'Converted sherpa-onnx Qwen3-ASR model with multilingual recognition and stronger long-form context handling.',

openless-all/app/src/i18n/es.ts

@@ -2110,6 +2110,9 @@ export const es: typeof zhCN = {
     sherpaModelWhisperLargeV3: 'Whisper Large V3 (multilingüe)',
     sherpaModelWhisperLargeV3Desc:
       'La versión multilingüe de código abierto más avanzada de Whisper: gran calidad y descarga de gran tamaño.',
+    sherpaModelZipformer: 'Zipformer en streaming (zh/en)',
+    sherpaModelZipformerDesc:
+      'Modelo en streaming de chino e inglés con la latencia más baja: el texto aparece mientras hablas.',
     sherpaModelQwen3: 'Qwen3-ASR 0.6B INT8',
     sherpaModelQwen3Desc:
       'Modelo Qwen3-ASR convertido para sherpa-onnx, con reconocimiento multilingüe y mejor tratamiento del contexto en textos largos.',

openless-all/app/src/i18n/fr.ts

@@ -2139,6 +2139,9 @@ export const fr: typeof zhCN = {
     sherpaModelWhisperLargeV3: 'Whisper Large V3 (multilingue)',
     sherpaModelWhisperLargeV3Desc:
       'La version multilingue open source la plus avancée de Whisper : grande qualité et téléchargement volumineux.',
+    sherpaModelZipformer: 'Zipformer streaming (zh/en)',
+    sherpaModelZipformerDesc:
+      'Modèle streaming chinois-anglais à la latence la plus faible — le texte apparaît pendant que vous parlez.',
     sherpaModelQwen3: 'Qwen3-ASR 0.6B INT8',
     sherpaModelQwen3Desc:
       'Modèle Qwen3-ASR converti pour sherpa-onnx, avec reconnaissance multilingue et meilleure gestion du contexte long.',

openless-all/app/src/i18n/ja.ts

@@ -2041,6 +2041,9 @@ export const ja: typeof zhCN = {
     sherpaModelWhisperLargeV3: 'Whisper Large V3（多言語）',
     sherpaModelWhisperLargeV3Desc:
       'オープンソース多言語モデルの中で最高品質の Whisper 系。高品質だが大容量。',
+    sherpaModelZipformer: 'Zipformer ストリーミング（中英）',
+    sherpaModelZipformerDesc:
+      '話しながら文字が出るストリーミング型の中英モデル。遅延が最も小さく、リアルタイム文字起こしに適します。',
     sherpaModelQwen3: 'Qwen3-ASR 0.6B INT8',
     sherpaModelQwen3Desc:
       '変換済み sherpa-onnx Qwen3-ASR モデル。多言語認識とより強い長文コンテキスト処理に対応。',

openless-all/app/src/i18n/ko.ts

@@ -2025,6 +2025,9 @@ export const ko: typeof zhCN = {
     sherpaModelWhisperLargeV3: 'Whisper Large V3 (다국어)',
     sherpaModelWhisperLargeV3Desc:
       '오픈소스 다국어 모델 중 품질이 가장 좋은 Whisper 계열. 고품질이지만 용량이 큽니다.',
+    sherpaModelZipformer: 'Zipformer 스트리밍(중·영)',
+    sherpaModelZipformerDesc:
+      '말하는 동안 텍스트가 나오는 스트리밍 중·영어 모델로, 지연이 가장 낮아 실시간 받아쓰기에 적합합니다.',
     sherpaModelQwen3: 'Qwen3-ASR 0.6B INT8',
     sherpaModelQwen3Desc:
       '변환된 sherpa-onnx Qwen3-ASR 모델로 다국어 인식과 더 강한 긴 문맥 처리를 지원합니다.',

openless-all/app/src/i18n/zh-CN.ts

@@ -1952,6 +1952,9 @@ export const zhCN = {
     sherpaModelWhisperLargeV3: 'Whisper Large V3（多语）',
     sherpaModelWhisperLargeV3Desc:
       '开源多语通用里效果最好的 Whisper 档，质量高、体积大，适合高质量转写。',
+    sherpaModelZipformer: 'Zipformer 流式（中英）',
+    sherpaModelZipformerDesc:
+      '边说边出的流式中英模型，延迟最低，适合实时听写。',
     sherpaModelQwen3: 'Qwen3-ASR 0.6B INT8',
     sherpaModelQwen3Desc:
       '转换后的 sherpa-onnx Qwen3-ASR 模型，支持多语言识别与更强的长上下文能力。',

openless-all/app/src/i18n/zh-TW.ts

@@ -1938,6 +1938,9 @@ export const zhTW: typeof zhCN = {
     sherpaModelWhisperLargeV3: 'Whisper Large V3（多語）',
     sherpaModelWhisperLargeV3Desc:
       '開源多語通用中效果最好的 Whisper 檔，品質高、體積大，適合高品質轉寫。',
+    sherpaModelZipformer: 'Zipformer 串流（中英）',
+    sherpaModelZipformerDesc:
+      '邊說邊出的串流中英模型，延遲最低，適合即時聽寫。',
     sherpaModelQwen3: 'Qwen3-ASR 0.6B INT8',
     sherpaModelQwen3Desc:
       '轉換後的 sherpa-onnx Qwen3-ASR 模型，支援多語言識別與更強的長上下文能力。',

openless-all/app/src/lib/localAsr.ts

@@ -37,7 +37,12 @@ export interface LocalAsrStorageSettings {
 export interface LocalAsrModelStatus {
   id: string;
   hfRepo: string;
+  displayName: string;
+  family: string;
+  mode: string | null;
+  languages: string[];
   downloadedBytes: number;
+  sizeBytes: number | null;
   isDownloaded: boolean;
 }
 
@@ -212,13 +217,23 @@ const MOCK_MODELS: LocalAsrModelStatus[] = [
   {
     id: 'qwen3-asr-0.6b',
     hfRepo: 'Qwen/Qwen3-ASR-0.6B',
+    displayName: 'Qwen3 ASR 0.6B',
+    family: 'qwen3_asr',
+    mode: null,
+    languages: ['zh', 'en', 'ja', 'ko', 'yue'],
     downloadedBytes: 0,
+    sizeBytes: null,
     isDownloaded: false,
   },
   {
     id: 'qwen3-asr-1.7b',
     hfRepo: 'Qwen/Qwen3-ASR-1.7B',
+    displayName: 'Qwen3 ASR 1.7B',
+    family: 'qwen3_asr',
+    mode: null,
+    languages: ['zh', 'en', 'ja', 'ko', 'yue'],
     downloadedBytes: 0,
+    sizeBytes: null,
     isDownloaded: false,
   },
 ];
@@ -302,6 +317,11 @@ export function deleteLocalAsrModel(modelId: string): Promise<void> {
   return invokeOrMock('local_asr_delete_model', { modelId }, () => undefined);
 }
 
+/** 清理中断下载遗留的 staging 目录，不触碰已安装模型。 */
+export function cleanupIncompleteLocalAsrModel(modelId: string): Promise<void> {
+  return invokeOrMock('local_asr_cleanup_incomplete', { modelId }, () => undefined);
+}
+
 export function getLocalAsrModelDir(modelId: string): Promise<string> {
   return invokeOrMock('local_asr_model_dir', { modelId }, () => '');
 }
@@ -422,6 +442,7 @@ export type SherpaOnnxModelAlias =
   | 'paraformer-zh'
   | 'whisper-small-multi'
   | 'whisper-large-v3-multi'
+  | 'zipformer-bilingual-zh-en-streaming'
   | 'qwen3-asr-0.6b-int8';
 
 export type SherpaOnnxMirror = 'huggingface' | 'hf-mirror' | 'github-release';
@@ -470,6 +491,11 @@ export const SHERPA_ONNX_ASR_MODELS: SherpaOnnxModelOption[] = [
     labelKey: 'localAsr.sherpaModelWhisperLargeV3',
     descKey: 'localAsr.sherpaModelWhisperLargeV3Desc',
   },
+  {
+    alias: 'zipformer-bilingual-zh-en-streaming',
+    labelKey: 'localAsr.sherpaModelZipformer',
+    descKey: 'localAsr.sherpaModelZipformerDesc',
+  },
   {
     alias: 'qwen3-asr-0.6b-int8',
     labelKey: 'localAsr.sherpaModelQwen3',
@@ -518,6 +544,13 @@ export function getSherpaOnnxAsrCatalog(): Promise<SherpaOnnxCatalogModel[]> {
       downloadedBytes: 0,
       fileSizeMb: 1700,
     },
+    {
+      alias: 'zipformer-bilingual-zh-en-streaming' as const,
+      displayName: 'Zipformer Streaming (zh/en)',
+      cached: false,
+      downloadedBytes: 0,
+      fileSizeMb: 260,
+    },
     {
       alias: 'qwen3-asr-0.6b-int8' as const,
       displayName: 'Qwen3-ASR 0.6B INT8',