Paul Guyot has put a lot of effort into building french Kaldi models using the Zamia Speech infrastructure. Quoting from https://github.com/pguyot/zamia-speech/releases :
Built using about 400 hours of audio, mostly licensed under CC BY-NC-SA, thus disallowing commercial use. The models themselves are licensed under CC BY-NC-SA 4.0.
TDNN 250 model took about two weeks of CPU (various VM sizes depending on steps) plus more than 48 hours with a Tesla K80. TDNN F model requires an additional 13 days with the same GPU.
%WER 30.17 [ 42464 / 140755, 3870 ins, 12822 del, 25772 sub ] exp/nnet3_chain/tdnn_250/decode_test/wer_7_0.0 %WER 25.40 [ 35749 / 140755, 3408 ins, 11742 del, 20599 sub ] exp/nnet3_chain/tdnn_f/decode_test/wer_7_0.0