Import custom ONNX speech models

README.md

@@ -120,6 +120,7 @@ The engine, model, and decoding cadence are **auto-selected to your machine** (c
 No mandatory engine downloads. No forced multi-gigabyte "summary engine" gating you before you can start. Wisp runs immediately, and **you** decide what to add:
 
 - **Pick any model** from the catalog — speed-first or accuracy-first — and switch per mode (Live vs File).
+- **Import your own speech models** — select a sherpa-onnx `.onnx` graph or Whisper GGML/GGUF file; Wisp validates it and automatically brings along matching tokens, encoder/decoder/joiner graphs, and external weights.
 - **Delete models** you don't need, right from the picker, to reclaim disk in one click.
 - **Configure everything** — transcription language, accuracy/speed profile, VAD gating, denoising, decoding thresholds, diarization, custom vocabulary — instead of being locked to one preset.
 - **Honest picker** — models your machine can't run are clearly marked, with size and hardware hints *before* you download.

README.zh-Hans.md

@@ -120,6 +120,7 @@ Wisp 不只是「能在 Mac 上跑」—— 它贴着架构做了优化：
 没有强制的引擎下载。没有逼你先下个几 GB 的「总结引擎」才能开始。Wisp 立即可用，**由你**决定加什么：
 
 - **任选模型** —— 从目录里挑，速度优先或精度优先 —— 还能按模式（实时 vs 文件）分别切换。
+- **导入自己的语音模型** —— 选择 sherpa-onnx `.onnx` 图或 Whisper GGML/GGUF 文件；Wisp 会先验证格式，并自动带上匹配的 tokens、encoder/decoder/joiner 图和外部权重。
 - **删除模型** —— 不需要的直接在选择器里一键删掉，腾回磁盘空间。
 - **一切可配** —— 转写语言、精度/速度档位、VAD 门控、降噪、解码阈值、说话人分离、自定义词表 —— 而不是被锁死在某个预设上。
 - **诚实的选择器** —— 你的机器跑不动的模型会被清楚标出，下载*之前*就给出体积和硬件提示。

app/src-tauri/Cargo.toml

@@ -54,6 +54,10 @@ tauri-plugin-dialog = "2"
 tauri-plugin-global-shortcut = "2"
 serde = { version = "1", features = ["derive"] }
 serde_json = "1"
+# Read and validate a user-selected ONNX graph plus its sherpa-onnx `model_type` metadata before
+# importing it. This exact version is already in the graph through wisp-embed, so it reuses the same
+# ONNX Runtime rather than adding a second native runtime.
+ort = { version = "=2.0.0-rc.9", features = ["download-binaries"] }
 
 wisp-core = { path = "../../crates/wisp-core" }
 # Insert dictated text into the frontmost app (clipboard + synthesized ⌘V). Stub off macOS.