storyteller-platform/citrinet512-de

Citrinet 512 German

Storyteller installs this model through ghost-story. This repository mirrors version 2.0.0 of citrinet512-de in the Storyteller model registry, with a folder per format.

Trained by neongeckocom/stt_de_citrinet_512_gamma_0_25, licensed BSD-3-Clause. Converted from the ONNX export OpenVoiceOS/neongeckocom-stt_de_citrinet_512_gamma_0_25_onnx at revision 6121e5c1bc2d.

• model.safetensors holds the encoder weights in the layout the MLX loader expects, plus the mel filterbank and window of the NeMo front end.
• model.onnx is the export with the mel front end spliced in front of the encoder by export-onnx.py, so it takes the pre-emphasised, reflect padded signal and computes the features and their length itself. The encoder is unchanged.
• config.json describes the architecture, the front end and the block layout, written by convert.py.
• vocab.txt is unchanged from the export.
• tokenizer.spm is the SentencePiece model from neongeckocom/stt_de_citrinet_512_gamma_0_25, onnx/tokenizer.spm.

The onnx format holds model.onnx, the mlx format model.safetensors, each with the configuration, vocabulary and tokenizer.

convert.py and export-onnx.py are in libraries/mlx-audio/scripts.