Swiss-German-Speech-Data---Meeting-Notes.md

@@ -13,10 +13,16 @@ Yixuan Xu, Daniel Perruchoud, Roberto Salomone, Fabian Jeitziner, Michael Graber
 
 * Performance of current Apertus checkpoints seems better than any other open model
 * SFT training paradigms / datasets
-  * https://huggingface.co/datasets/gpt-omni/VoiceAssistant-400K 
+  * https://huggingface.co/datasets/gpt-omni/VoiceAssistant-400K
   * https://huggingface.co/datasets/Harland/AudioMCQ-StrongAC-GeminiCoT
   * https://huggingface.co/datasets/AIDC-AI/Marco_Longspeech
-* 
+* Apertus 2.0 pre-training is planned for July
+* Yixuan: VoxCPM2 seems better than CosyVoice
+
+### New Action Points
+
+* [ ] FHNW: share synthesized SFT QA questions via huggingface
+* [ ] FHNW: evaluate synthesized SFT audio quality via ASR 
 
 ## Meeting May 13, 2026