POCKETSPHINX_LM_EVAL(1)

📦 version 5.1.1, in nixpkgs unstable, ubuntu resolute, ubuntu stonking

✨ pocketsphinx_lm_eval ✨

okay bestie, so this is a page from the General Commands Manual, and it's literally about a little tool called pocketsphinx_lm_eval — and no, you don't need to be a linguistics girlie to get it 💅

🎀 NAME 🎀

pocketsphinx_lm_eval — evaluate perplexity of a transcription

like. it grades how confused a language model is by your text. that's the whole job. iconic.

🌸 SYNOPSIS 🌸

pocketsphinx_lm_eval [ options ]...

that's it. options only. no positional args, no drama 🦋

💖 DESCRIPTION 💖

So here's the vibe: this program evaluates the perplexity of a text file according to a given language model. Perplexity = the "how shook is the model by this sentence" number. Lowkey the whole point of running this thing.

The text file is assumed to be in transcript format, meaning: one utterance per line, delimited by <s> and </s>. So every line gets its own little <s> ... </s> hug, fr fr. 🫶

Now the options, and yes we're doing all of them, no skipping:

  • -help — Shows the usage of the tool. Your classic "wait what do I type again" button.
  • -lm model file — the model file. your LM. the moment.
  • -lmctlfn — file listing a set of language models. a whole roster, like a group chat of LMs.
  • -lmname — of language model in -lmctlfn to use for all utterances. pick your main character from the roster ✨
  • -logbase — Base in which all log-likelihoods calculated. math girlies, this one's for you 💜
  • -lsn file to evaluate — file to evaluate. yes, the text file from before.
  • -lw — Language model weight. how much say the LM gets.
  • -mmap — Use memory-mapped I/O for reading binary LM files. fancy and fast, no cap.
  • -probdef — definition file for classes in LM.
  • -text string to evaluate — string to evaluate. skip the file, just hand it a string, diva behavior 💅
  • -uw — Unigram probability weight (interpolated with uniform distribution). that's the whole description, I'm just the messenger 🍓
  • -verbose — Print details of perplexity calculation. not me wanting to see the receipts 😭
  • -wip — Word insertion probability.

👑 AUTHOR 👑

David Huggins-Daines <dhdaines@gmail.com> — king behavior, honestly 🫶

Copyright © 2007-2008 Carnegie Mellon University. See the file COPYING included with this package for more information.

real talk: that COPYING file is not a vibe, it's a legal document. go read it if you're doing anything with this.

and that's the whole page bestie!! 💖 the source is the General Commands Manual, version vibes of pocketsphinx_lm_eval(1), dated 2008-05-12 — an absolute elder of a manpage, still slaying. mwah 💋

📦 where this page comes from