docs(audio-lab): README updates for ASR (M3)

README.md

@@ -18,7 +18,7 @@ A research repo for the [WaveKat](https://github.com/wavekat) project — intera
 ```
 wavekat-lab/
 ├── tools/
-│   ├── audio-lab/     Real-time VAD + Turn Detection comparison app (Rust + React)
+│   ├── audio-lab/     Real-time VAD + Turn Detection + ASR comparison app (Rust + React)
 │   └── cv-explorer/   Mozilla Common Voice dataset browser (Cloudflare Workers + React)
 ├── notebooks/         Jupyter notebooks (training, validation, dataset splits)
 └── docs/              Plans and design docs
@@ -30,9 +30,9 @@ Each tool is self-contained — its own Makefile, lockfiles, and build setup liv
 
 ### [Audio Lab](tools/audio-lab/) — `tools/audio-lab/`
 
-Web app for testing and comparing WaveKat library backends side by side in real time. Live mic capture, WAV upload, multi-config fan-out, VAD-gated pipeline mode, waveform + spectrogram + probability timelines.
+Web app for testing and comparing WaveKat library backends side by side in real time. Live mic capture, WAV upload, multi-config fan-out, VAD-gated pipeline mode, live ASR transcripts, waveform + spectrogram + probability timelines.
 
-Backends: webrtc-vad, silero-vad, ten-vad, firered-vad, pipecat smart-turn. [Details →](tools/audio-lab/README.md)
+Backends: webrtc-vad, silero-vad, ten-vad, firered-vad, pipecat smart-turn, sherpa-onnx ASR. [Details →](tools/audio-lab/README.md)
 
 ### [Common Voice Explorer](tools/cv-explorer/) — `tools/cv-explorer/`
 

tools/audio-lab/Makefile

@@ -1,5 +1,5 @@
 SHELL := /bin/bash
-NVM   := source $$NVM_DIR/nvm.sh && nvm use --silent &&
+NVM   := source $$NVM_DIR/nvm.sh &&
 
 .PHONY: help \
         install install-backend install-frontend \
@@ -38,15 +38,15 @@ install-backend:
 	cargo install cargo-watch
 
 install-frontend:
-	$(NVM) cd frontend && npm install
+	$(NVM) cd frontend && nvm use --silent && npm install
 
 # ─── Development ──────────────────────────────────────────────────────────────
 
 dev-backend:
 	cargo watch -x 'run -p audio-lab'
 
 dev-frontend:
-	$(NVM) cd frontend && npm run dev
+	$(NVM) cd frontend && nvm use --silent && npm run dev
 
 dev:
 	@echo "Run 'make dev-backend' and 'make dev-frontend' in separate terminals"

tools/audio-lab/README.md

@@ -1,6 +1,6 @@
 # Audio Lab
 
-A web-based experimentation tool for testing and comparing [WaveKat](https://github.com/wavekat) library backends — VAD, turn detection, and more — side by side in real time.
+A web-based experimentation tool for testing and comparing [WaveKat](https://github.com/wavekat) library backends — VAD, turn detection, ASR, and more — side by side in real time.
 
 > [!WARNING]
 > Early development. Things may change.
@@ -10,6 +10,7 @@ A web-based experimentation tool for testing and comparing [WaveKat](https://git
 - **Live recording** — capture audio from your microphone server-side, stream results to the browser in real time
 - **File analysis** — upload a WAV file and run multiple configs against it at full speed
 - **Side-by-side comparison** — fan out audio to N configurations simultaneously and compare outputs
+- **Live transcripts** — stream partial + final ASR transcripts per config as audio plays
 - **Preprocessing exploration** — apply high-pass filters, RNNoise denoising, or normalization per-config
 - **Interactive visualization** — waveform, spectrogram, and probability timelines with synchronized zoom, pan, and hover
 
@@ -57,6 +58,17 @@ Each config can also enable per-config preprocessing: high-pass filter, RNNoise
 |---------|-------------|-------|
 | **pipecat** | Pipecat Smart Turn v3 — audio-based EOU detection | 16 kHz PCM audio |
 
+### ASR
+
+Streaming speech-to-text via [`wavekat-asr`](https://github.com/wavekat/wavekat-asr). Each ASR config gets a per-config transcript card stacked under the timelines: committed finals with `[mm:ss.s–mm:ss.s]` timestamps, a dimmed trailing line for the live partial, and footer stats (last confidence, count, average segment duration).
+
+| Backend | Description | Preset (`preset` param) |
+|---------|-------------|-------------------------|
+| **sherpa-onnx** | Local streaming Zipformer / Paraformer via [sherpa-onnx](https://github.com/k2-fsa/sherpa-onnx) | `bilingual` (default, ZH+EN) · `en` · `zh` · `paraformer-zh-en` |
+
+> [!NOTE]
+> The first time you record or load a file with an ASR config enabled, sherpa-onnx downloads the chosen model from HuggingFace (~75 MB for `bilingual`) into `$HF_HOME` (default `~/.cache/huggingface/hub/`). The transcript card shows `loading model…` until the model is ready; subsequent runs are instant.
+
 ## Architecture
 
 The Rust backend handles all audio capture and processing; the React frontend is embedded in the binary and handles visualization only.