WWW
While working with phonemes at the sentence level it is hard to compose word sequences out of those phonemes. This ticket would need to investigate methods/papers/online toolkits where phonemes can be mapped back to words. Traditionally ASR systems used a lexicon like the CMUDict Lexicon or use a finite state transducer (FST). Our G2P is an FST under the hood, so that mapping might be tool specific. There might be a mode within our choice of G2P to convert phoneme sequences back to words. I may be that we might need to use an FST based on OpenFST via Kaldi or some other tool.
AC
A wiki page on possible alternatives and options.
jindaznb
commented
8 months ago
WWW While working with phonemes at the sentence level it is hard to compose word sequences out of those phonemes. This ticket would need to investigate methods/papers/online toolkits where phonemes can be mapped back to words. Traditionally ASR systems used a lexicon like the CMUDict Lexicon or use a finite state transducer (FST). Our G2P is an FST under the hood, so that mapping might be tool specific. There might be a mode within our choice of G2P to convert phoneme sequences back to words. I may be that we might need to use an FST based on OpenFST via Kaldi or some other tool.
AC A wiki page on possible alternatives and options.