v0.2.0 - Cyberistic Full-Mixed Text CTC

Summary

• Promotes Cyberistic's full-mixed FastConformer text CTC runtime as the current release path.
• Ships fastconformer_full_mixed.onnx plus the BPE vocab, Quran CTC token table, tokenizer metadata, export metadata, and Quran text needed for offline inference.
• Keeps prior q8/phoneme release assets historical; new integrations should use raw 16 kHz audio input with audio_signal and length tensors.

Assets

• fastconformer_full_mixed.onnx - 88 MB mixed int4/int8 ONNX, raw waveform input
• vocab.json - 1025-token Arabic BPE CTC vocabulary
• quran_ctc_tokens.json - precomputed verse/span token IDs for CTC reranking
• tokenizer.model - tokenizer provenance artifact
• export_metadata.json - hashes and runtime metadata
• quran.json - Quran verse text used by the matcher

v0.1.0 - Browser ONNX Runtime

Quantized ONNX model (uint8, 132MB) for browser-based inference via ONNX Runtime Web.