Expand model card

README.md

@@ -6,27 +6,53 @@ tags:
 - speech
 - gemma-4
 - csm
+- lora
 ---
 
 # Parley
 
-A full-duplex speech model: Gemma 4 12B listens while it talks and decides every 80 ms
-whether to speak, keep speaking, or listen. CSM-1B speaks its words.
+Parley is a full-duplex speech model. It listens while it talks and decides every 80 ms
+whether to speak, keep speaking, or listen. There is no turn detector: the model itself
+makes every timing decision, frame by frame.
 
-This repository contains the LoRA adapter and processor for the `e498` checkpoint
-(`talker-v6`, seed 43), the live model used through a noise gate.
+## What it is
 
-## Usage
+- **Thinker:** Gemma 4 12B (audio-native, no speech encoder). It hears 16 kHz audio in
+  40 ms slices and writes text or control tokens on an 80 ms clock: `SIL` while listening,
+  `EPS` while still speaking, or words when it has them.
+- **Talker:** CSM-1B (Sesame) speaks the words Gemma writes, in chunks, through a playout
+  queue that handles interruption and yielding.
+- **Codec:** Mimi (Kyutai) provides the audio codes CSM renders.
 
-```bash
-pip/uv install from https://github.com/walln/parley
-uv run parley download walln/parley --out checkpoints/parley
-uv run parley serve --checkpoint checkpoints/parley --prompt "You are a helpful assistant."
-```
+Gemma never predicts audio. The thinker/talker split keeps recall high and avoids the
+acoustic glitches of a single model generating its own codes.
 
-## License
+## Intended use
 
-Weights: research-only (CC-BY-NC 4.0). Code: Apache-2.0.
+Research, experimentation, and demos of full-duplex voice interaction. Not a production
+voice agent. See the project repo for the live demo and code:
+https://github.com/walln/parley
 
-Base components: Gemma 4 12B (Apache 2.0), CSM-1B (Apache 2.0, Llama 3.2
-Community License), Mimi (CC-BY 4.0).
+## Limitations
+
+- Trained on synthetic voices only; recorded human speech is weaker.
+- One voice (the CSM reference clip).
+- Gemma gets one text token per frame, so long text/tool calls do not fit.
+- Research-only weights.
+
+## Components and licenses
+
+- Gemma 4 12B (`google/gemma-4-12B-it`) — Apache 2.0.
+- CSM-1B (`sesame/csm-1b`) — Apache 2.0, built on Llama 3.2 (Llama Community License).
+- Mimi (`kyutai/mimi`) — CC-BY 4.0.
+
+## Weights license
+
+Research-only (CC-BY-NC 4.0). Commercial use requires a separate agreement.
+
+## Citation
+
+Blog post: "Inside Frankenstein's lab: a full-duplex model at home"
+(https://nickwall.dev, 2026). Code: https://github.com/walln/parley.
+EOF
+echo done