Show four MP4 demos and expand section-level documentation links

README.md

@@ -34,42 +34,42 @@ into a variable set of source maps. The release includes the complete inference,
 evaluation, training, and Online Scene Generation workflows described in the
 paper.
 
-## Demos
+Details: [Model architecture](docs/model_architecture.md).
 
-The following 20-second examples use the SAID (PaSST) paper checkpoint. Ground
-Truth is shown on the left and Prediction on the right with a shared timeline,
-class legend, colors, and display scale. Each MP4 includes synchronized audio;
-the inline animation previews the visual output.
+## Demos
 
-### Demo 1 — TAU scene
+The following 20-second examples show SAID (PaSST) and Audio2Sph + Panoramic
+Decoder outputs. Ground Truth is shown on the left and Prediction on the right
+with a shared timeline, colors, and display scale. The SAID videos also include
+the sound-event class legend. All four MP4 files include synchronized audio.
 
-[![SAID Demo 1: Ground Truth and Prediction in a TAU scene](docs/media/said_demo_1.webp)](docs/media/said_demo_1.mp4)
+### SAID (PaSST) — TAU scene
 
-[Open Demo 1 as MP4](docs/media/said_demo_1.mp4)
+<video src="https://huggingface.co/IN03X/SAID/resolve/main/docs/media/said_demo_1.mp4" controls playsinline preload="metadata" width="100%"></video>
 
-### Demo 3 — Sony scene
+### SAID (PaSST) — Sony scene
 
-[![SAID Demo 3: Ground Truth and Prediction in a Sony scene](docs/media/said_demo_3.webp)](docs/media/said_demo_3.mp4)
+<video src="https://huggingface.co/IN03X/SAID/resolve/main/docs/media/said_demo_3.mp4" controls playsinline preload="metadata" width="100%"></video>
 
-[Open Demo 3 as MP4](docs/media/said_demo_3.mp4)
+### Audio2Sph + Panoramic Decoder — TAU scene
 
-<details>
-<summary>Audio2Sph + Panoramic Decoder demonstrations</summary>
+<video src="https://huggingface.co/IN03X/SAID/resolve/main/docs/media/audio2sph_demo_1.mp4" controls playsinline preload="metadata" width="100%"></video>
 
-These class-agnostic examples visualize the panoramic acoustic field used during
-Audio2Sph pretraining.
+### Audio2Sph + Panoramic Decoder — Sony scene
 
-- [TAU scene](docs/media/audio2sph_demo_1.mp4)
-- [Sony scene](docs/media/audio2sph_demo_3.mp4)
+<video src="https://huggingface.co/IN03X/SAID/resolve/main/docs/media/audio2sph_demo_3.mp4" controls playsinline preload="metadata" width="100%"></video>
 
-</details>
+Details: [Demos and inference](docs/inference.md).
 
 ## Quick start
 
-SAID supports Python 3.10 through 3.12. Install the project from the repository
-root:
+SAID supports Python 3.10 through 3.12. Create an environment, enter the project
+directory, and install the package:
 
 ```bash
+conda create -n said python=3.10 -y
+conda activate said
+cd /path/to/SAID
 pip install -e .
 ```
 
@@ -90,8 +90,9 @@ The CLI downloads the selected checkpoint from
 `checkpoints/` directory on first use. `recording.wav` may contain the complete
 32-channel Eigenmike signal or the four selected capsules `[6, 10, 26, 22]` in
 that order. The default output is readable DCASE JSON. Select the AudioMAE
-variant with `--model said_audiomae`; complete options and the Python API are in
-[Inference](docs/inference.md).
+variant with `--model said_audiomae`.
+
+Details: [Inference and Python API](docs/inference.md).
 
 ## Results
 
@@ -120,8 +121,8 @@ The table evaluates all 78 full development-test recordings.
 
 The `(4.2)` AP columns reproduce the evaluator revision used by the paper;
 `(6.30)` applies the current official evaluator to the same predictions.
-[Evaluation](docs/evaluation.md#metric-versions) records the exact dates,
-commits, definitions, and output fields.
+
+Details: [Metric versions and definitions](docs/evaluation.md#metric-versions).
 
 ## Evaluation
 
@@ -140,6 +141,8 @@ The default metrics follow the current `(6.30)` official definition;
 The evaluator is acquired from the
 [official DCASE2026 Task 3 repository](https://github.com/iranroman/DCASE2026_Task3_SAISELD_baseline).
 
+Details: [DCASE evaluation and reproduction](docs/evaluation.md).
+
 ## Models and checkpoints
 
 | Model | Role and output | Checkpoint |
@@ -153,6 +156,10 @@ and SHA256 before loading. They are released for non-commercial research use onl
 under the [SAID Model Weights Non-Commercial Research License 1.0](LICENSES/SAID-Model-Weights-NonCommercial-1.0.txt)
 and all applicable upstream terms.
 
+Details: [SAID (PaSST)](docs/checkpoints/said_passt.md),
+[SAID (AudioMAE)](docs/checkpoints/said_audiomae.md), and
+[Audio2Sph + Panoramic Decoder](docs/checkpoints/audio2sph.md).
+
 ## Training and data
 
 The paper training workflow is:
@@ -179,9 +186,8 @@ the included adapters:
 - VCTK v0.80 for Audio2Sph pretraining;
 - an authorized SourceBank manifest for class-labeled Online Scene Generation.
 
-[Training](docs/training.md) gives the complete checkpoint flow and recipes.
-[Data preparation](docs/data.md) gives the required sources, directory layouts,
-manifest schema, and license-aware interfaces.
+Details: [Training](docs/training.md), [data preparation](docs/data.md), and
+[configuration](docs/configuration.md).
 
 ## Documentation