ASR decoder written in Common Lisp. Uses acoustic models trained by Sphinx, and ARPA LM files.
This may not currently be fully functional, but it's probably close.
This can be run in batch mode on Sphinx feature extracted files. The 'standalone' component operates on raw WAV files and does the feature extraction internally.
Run from the command line args similar to Sphinx3 decode:
scripts/decode.sh
Use this function to Load a Sphinx3 model:
(defun load-s3-model (&key folder mdef mean var mixw tmat lda dict))You can either specify a folder containing all the model file (with standard names) or specify each model file individually.
Here's an example of loading models from a folder:
(defvar *am* (cl-asr::load-s3-model :folder "/Users/bel/workspace/sphinx/acoustic-models/fisher/"))The language model gets loaded like so:
(defvar *lm* (cl-lm::load-model <filename>))