PhoneticXeus Multilingual phone recognition that turns speech into IPA phones, built on the XEUS speech encoder. Trained on 70+ languages (IPAPack++). 🎤 Try it in your browser: demo Space 💻 Code & training recipe: GitHub 📄 Paper: arXiv 2603.29042 Usage model.transcribe(...) returns a list of dicts with processed transcript (joined IPA) and predicted transcript (slash separated phones). Calling model(input values) returns frame level CTC logits (batch, frames, 428) for custom decoding. Audio must be mono 16 kHz . The first load asks you to allow the repo's remote code ( trust remote code=True ). Citation
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy