gtcrn-stream

Verbatim mirror of the upstream GTCRN streaming ONNX model from Xiaobin-Rong/gtcrn (stream/onnx_models/gtcrn_simple.onnx, trained on DNS3).

Re-hosted here so it resolves through the same HF Hub cache / HF_HUB_OFFLINE=1 path as every other model in the uri/rei stack (mic_capture [denoise.model]).

• gtcrn_simple.onnx — simplified streaming graph, 16 kHz native, ~24 K params. Inputs: mix [1,257,1,2], conv_cache [2,1,16,16,33], tra_cache [2,3,1,1,16], inter_cache [2,1,33,16]. Outputs: enh + *_out caches. STFT: n_fft 512, hop 256, sqrt-Hann window.

License: MIT (upstream).