Set helpful configuration settings for the phonetic recognizer

Copied from the PocketSphinx recognizer
This commit is contained in:
Daniel Wolf 2019-01-21 22:33:34 +01:00
parent c821788569
commit 06b4855d6d
1 changed files with 7 additions and 0 deletions

View File

@ -25,6 +25,13 @@ static lambda_unique_ptr<ps_decoder_t> createDecoder(optional<std::string> dialo
// Low values (<= 0.4) can lead to fluttering animation. // Low values (<= 0.4) can lead to fluttering animation.
// High values (>= 1.0) can lead to imprecise or freezing animation. // High values (>= 1.0) can lead to imprecise or freezing animation.
"-lw", "0.8", "-lw", "0.8",
// Add noise against zero silence
// (see http://cmusphinx.sourceforge.net/wiki/faq#qwhy_my_accuracy_is_poor)
"-dither", "yes",
// Disable VAD -- we're doing that ourselves
"-remove_silence", "no",
// Perform per-utterance cepstral mean normalization
"-cmn", "batch",
// The following settings are recommended at // The following settings are recommended at
// http://cmusphinx.sourceforge.net/wiki/phonemerecognition // http://cmusphinx.sourceforge.net/wiki/phonemerecognition