feat(asr): add pluggable ONNX streaming backend and model registry

Add a second backend (sherpa-onnx) alongside llama.cpp, so native streaming models can be selected by ID from a JSON registry. All such models share one adapter: only the model files and a few options differ. Models now live under the app data directory and are fetched by scripts/fetch-models.sh. Adding a model is a data change, not code. Measured on the same audio: the streaming backend is ~35x cheaper in CPU than Qwen3-ASR (0.10-0.12 vs 4.27 core-seconds per audio second), shows the first caption at 0.48-1.44s, and matches its accuracy for zh/en. 新增 sherpa-onnx 流式后端，与 llama.cpp 并存；模型按注册表 id 从应用数据目录加载。同族模型共用一个适配器，新增模型只改 data/models.json。实测流式后端 CPU 约为 Qwen3-ASR 的 1/35，首个字幕 0.48~1.44s，中英准确率持平。注意预编译库是旧 C++ ABI，必须走其 C API。
2026-10-12 16:06:51 +08:00
1 parent 6d64987d64
commit bd921b1ac5
@@ -0,0 +1,88 @@
+{
+  "version": 1,
+  "comment": "live-caption 的模型注册表。新增模型 = 在这里加一条，不需要改代码。",
+  "models": [
+    {
+      "id": "zipformer-zh-2025",
+      "name": "中文流式（k2-fsa 官方 2025-06-30）",
+      "backend": "sherpa-online",
+      "languages": ["zh"],
+      "license": "Apache-2.0",
+      "note": "154MB，实测中文逐字正确；首个字幕 0.48s，约 0.10 核·秒/秒音频",
+      "files": {
+        "encoder": {
+          "name": "encoder.int8.onnx",
+          "url": "https://hf-mirror.com/csukuangfj/sherpa-onnx-streaming-zipformer-zh-int8-2025-06-30/resolve/main/encoder.int8.onnx"
+        },
+        "decoder": {
+          "name": "decoder.onnx",
+          "url": "https://hf-mirror.com/csukuangfj/sherpa-onnx-streaming-zipformer-zh-int8-2025-06-30/resolve/main/decoder.onnx"
+        },
+        "joiner": {
+          "name": "joiner.int8.onnx",
+          "url": "https://hf-mirror.com/csukuangfj/sherpa-onnx-streaming-zipformer-zh-int8-2025-06-30/resolve/main/joiner.int8.onnx"
+        },
+        "tokens": {
+          "name": "tokens.txt",
+          "url": "https://hf-mirror.com/csukuangfj/sherpa-onnx-streaming-zipformer-zh-int8-2025-06-30/resolve/main/tokens.txt"
+        }
+      },
+      "options": {}
+    },
+    {
+      "id": "xasr-zh-en-480",
+      "name": "中英流式（X-ASR-zh-en, 480ms）",
+      "backend": "sherpa-online",
+      "languages": ["zh", "en"],
+      "license": "Apache-2.0",
+      "note": "587MB，实测中英准确率与 Qwen3-ASR 持平；首个字幕 1.44s",
+      "files": {
+        "encoder": {
+          "name": "encoder-480ms.onnx",
+          "url": "https://hf-mirror.com/GilgameshWind/X-ASR-zh-en/resolve/main/deployment/models/chunk-480ms-model/encoder-480ms.onnx"
+        },
+        "decoder": {
+          "name": "decoder-480ms.onnx",
+          "url": "https://hf-mirror.com/GilgameshWind/X-ASR-zh-en/resolve/main/deployment/models/chunk-480ms-model/decoder-480ms.onnx"
+        },
+        "joiner": {
+          "name": "joiner-480ms.onnx",
+          "url": "https://hf-mirror.com/GilgameshWind/X-ASR-zh-en/resolve/main/deployment/models/chunk-480ms-model/joiner-480ms.onnx"
+        },
+        "tokens": {
+          "name": "tokens.txt",
+          "url": "https://hf-mirror.com/GilgameshWind/X-ASR-zh-en/resolve/main/deployment/models/chunk-480ms-model/tokens.txt"
+        }
+      },
+      "options": {}
+    },
+    {
+      "id": "nemotron35-560ms",
+      "name": "多语言流式（Nemotron 3.5, 560ms, 含日语）",
+      "backend": "sherpa-online",
+      "languages": ["en", "ja", "ko", "vi", "ru", "ar", "hi", "fr", "de", "es", "it", "pt", "nl", "tr", "uk", "zh"],
+      "license": "OpenMDW-1.1",
+      "note": "651MB，40 种语言 + 内置语言识别；实测中英日都可用但都有瑕疵", 
+      "optionsHint": "language",
+      "files": {
+        "encoder": {
+          "name": "encoder.int8.onnx",
+          "url": "https://hf-mirror.com/csukuangfj2/sherpa-onnx-nemotron-3.5-asr-streaming-0.6b-560ms-int8-2026-06-11/resolve/main/encoder.int8.onnx"
+        },
+        "decoder": {
+          "name": "decoder.int8.onnx",
+          "url": "https://hf-mirror.com/csukuangfj2/sherpa-onnx-nemotron-3.5-asr-streaming-0.6b-560ms-int8-2026-06-11/resolve/main/decoder.int8.onnx"
+        },
+        "joiner": {
+          "name": "joiner.int8.onnx",
+          "url": "https://hf-mirror.com/csukuangfj2/sherpa-onnx-nemotron-3.5-asr-streaming-0.6b-560ms-int8-2026-06-11/resolve/main/joiner.int8.onnx"
+        },
+        "tokens": {
+          "name": "tokens.txt",
+          "url": "https://hf-mirror.com/csukuangfj2/sherpa-onnx-nemotron-3.5-asr-streaming-0.6b-560ms-int8-2026-06-11/resolve/main/tokens.txt"
+        }
+      },
+      "options": {}
+    }
+  ]
+}