I have hooked up the hub4 acoustic and language models which I obtained from the sphinx web site to sphinx 3.5
The accuracy with my homebrew test data consisting of American speakers culled from the web (can't afford real test data) is rather variable - 100% to 0%. So I wondered whether it would be possible to get hold of the configuration file used by the CMU team to benchmark sphinx 4 and sphinx 3.5 (or similar) using your HUB4 models?
One final question - sphinx 4 calculates word error rates and sentence error rates - presumably I would need to get hold of SCLITE or similar to score my test set for s3.5 or am I missing something?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I have hooked up the hub4 acoustic and language models which I obtained from the sphinx web site to sphinx 3.5
The accuracy with my homebrew test data consisting of American speakers culled from the web (can't afford real test data) is rather variable - 100% to 0%. So I wondered whether it would be possible to get hold of the configuration file used by the CMU team to benchmark sphinx 4 and sphinx 3.5 (or similar) using your HUB4 models?
One final question - sphinx 4 calculates word error rates and sentence error rates - presumably I would need to get hold of SCLITE or similar to score my test set for s3.5 or am I missing something?