kaldi-gstreamer-server VS ChessPositionRanking

Through my research, the most promising real-time transcription options appear to be Vosk or Kaldi Gstreamer. I’ve set them both up & they appear to work well for general transcription, but I’m not sure how to handle the user-specific custom vocabularies.

It is kind of difficult to find something like this free of charge (and open source) since the ASR service needs to be hosted somewhere. If you are really interested in the topic then you could take a lit into kaldi and its pretrained models (but kaldi is kind of difficult to learn so I don't really recommend it if you want something quick) and then you could also combine that with kaldi-gstreamer in order to set up a server which you can turn on and off whenever you like.

3. There's extra nuanced things you might want to handle in the coding, like that pawns can't be on their own back row. That is significantly harder.

It looks to me like https://github.com/tromp/ChessPositionRanking has resolved these sorts of issues, but I haven't dug into exactly how.

Multinomial rankings can be combined with a dozen others to rank a subset of all chess positions including all legal ones. This allows one to sample millions of random such positions, determine how many are legal, and thus obtain an accurate estimate of 4.8&10^44 legal chess positions [2].

[1] https://github.com/tromp/ChessPositionRanking/blob/main/src/...

[2] https://github.com/tromp/ChessPositionRanking

The key words being "without promotion". Both bounds, this one and Steinerberger's, only consider positions reachable without promotion. Allowing promotions, one estimate suggests that the number is close to 4.82 × 10^44.

source

This one as well as Shannon number wiki seem to say that possible sensible moves are about 10^40 while and 10^120 while taking any moves (maybe including some illogical / illegal ones) .

Shannon's estimate was based on very primitive methods; by generating random positions and using fairly advanced methods to see whether they are legal or not (ie., can you construct a proof game for it, or prove that it could never happen), you will get much closer. A group of people have been working on this, and their current best estimate is (4.822 +- 0.028) * 10^44, or a bit over 148 bits. (Amazingly enough, Shannon wasn't all that far off on this account! His estimated number of legal games seems much more dodgy, though.)

http://talkchess.com/forum3/viewtopic.php?f=7&t=77685&sid=e3...

Practically speaking, https://github.com/tromp/ChessPositionRanking gives a number between 0 and approx. 8.7 * 10^45 for any legal position, so it's only a couple of bits away from optimality.

The number of chess positions has now been estimated with 2 digits of accuracy as ~ 4.8 x 10^44: https://github.com/tromp/ChessPositionRanking