Run phonetic language model for word level ASR

[aanchan]

WWW As a researcher wanting to understand the impact of phonetic error correction using language models on word-level recognition from dysarthric speakers I would like to run an experiment using acoustic models trained on TORGO. Acoustic models are trained on a per-speaker basis on two different scenarios.

• The results for the two training scenarios on TORGO are here: https://github.com/SlangLab-NU/torgo_inference_on_cluster
• [ ] A previously trained phonetic language model and its use in speech recognition is here : https://github.com/SlangLab-NU/links (Look at PSST with LM)
• [ ] For a G2P tool that allows for P2G we might want to look at CMU's phoneme to grapheme and grapheme to phoneme tool for inverse G2P (https://github.com/cmusphinx/g2p-seq2seq)

AC A set of experiments on word level ASR output with phoneme level correction

[jindaxz]

Results for experiments: https://docs.google.com/spreadsheets/d/102zO3a8PhgyPbgwtdlbRDtYEBC4lorDzuhVEiKaKkIw/edit#gid=0