Upload README.md with huggingface_hub

README.md

@@ -1,3 +1,19 @@
+---
+language:
+- en
+tags:
+- turn-taking
+- voice-activity-detection
+- speech
+- conversation
+- audio
+- dualturn
+- endpointing
+base_model:
+- anyreach-ai/dualturn-qwen2.5-mimi-0.5B
+license: apache-2.0
+---
+
 # anyreach-ai/dualturn-endpointing
 
 Real-time speech endpoint detector for two-channel (user + agent) audio.
@@ -116,27 +132,6 @@ for chunk in audio_source():
 
 ---
 
-## Performance (held-out test set, 15 calls)
-
-| Metric | Value |
-|--------|-------|
-| ST recall | 99% |
-| ST precision | 90% |
-| AUC (ST vs CL) | 0.853 |
-| Threshold | 0.30 |
-
-Tuned for **high ST recall** — almost never misses a real turn end. Accepts some false STs to avoid making the user repeat themselves.
-
----
-
-## Training Data
-
-Endpoint classifier trained on 70 real dual-channel calls (user + Gemini agent) automatically labelled by Gemini 2.5 Pro.
-
-Backbone: [anyreach-ai/dualturn-qwen2.5-mimi-0.5B](https://huggingface.co/anyreach-ai/dualturn-qwen2.5-mimi-0.5B)
-
----
-
 ## Authors
 
 * Shangeth Rajaa — Senior ML Research Scientist, Anyreach AI