|
SPRAAK
|
All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Groups Pages
|
SPRAAK
|
Collaboration diagram for SprCwrSearchStd:Data Fields | |
| unsigned int | load_flags |
| which parts (phone_set, lexicon, hmm, ...) are loaded More... | |
| unsigned int | modif_must_do |
| combine reset and free operations during modifications More... | |
| unsigned int | modif_have_done |
| combine reset and free operations during modifications More... | |
| char | store_seq |
| do (not) store the node/phoneme sequences during the search More... | |
| char | recog_mode |
| recognition mode More... | |
| char | lm_fwd_mode |
| type of LM-forwarding that is used More... | |
| char | sent_end_flags |
| for sentence/word end in CWR_best_sentence() More... | |
| char | wexpand_flags |
| word expand flags More... | |
| char | sdet_avail |
| the preprocessing gives output on sentence boundaries More... | |
| char | sdet_cont |
| char | sdet_new_res |
| char | gauss_sel |
| gaussian selection status More... | |
| int | sdet_count |
| int | sdet_out |
| SprCwrWordSet | sentence_begin |
| word index/indices of <s> More... | |
| SprCwrWordSet | sentence_end |
| word index/indices of </s> More... | |
| float | lm_A |
| float | lm_C |
| upgrade the language model cost: cost = LM_cost*A+C More... | |
| unsigned int | lm_cache_size |
| target size of the LM-cache More... | |
| unsigned int | lmc_hash_size |
| target size of the LM-context hashing More... | |
| float * | slmsp |
| float * | slmsp0 |
| the unshifted unigram prob array More... | |
| int * | slmsp_wset_mx |
| word with best unigram probability in a word set More... | |
| int * | fwd_comb_inst |
| definition of word sets when using full LM-forwarding More... | |
| int ** | fwd_icomb_inst |
| instructions for a specific word set More... | |
| int | fwd_wset_len |
| number of new word sets introduced for full LM-forwarding More... | |
| int | Ntokens_max |
| maximal beam width (absolute limit on the nr. of open token) More... | |
| int | Ntokens_min |
| minimal beam width (absolute limit on the nr. of open token) More... | |
| int | Ntokens_wstart |
| int | Ntokens_wend |
| int | Ntokens_wint |
| int | Nctokens_wend |
| int | Nctokens_wint |
| unsigned int | beam_corr_up |
| unsigned int | beam_corr_down |
| unsigned long long | Ntokens_av_ax |
| maximum beam size after word end expansion More... | |
| unsigned long long | Ntokens_av_bp |
| maximum beam size before pruning More... | |
| unsigned long long | Ntokens_av_ap |
| maximum beam size after pruning More... | |
| double | av_threshold |
| average pruning threshold (adaptive) More... | |
| unsigned int | frame_cnt |
| int | iframe |
| number of the current frame More... | |
| int | iframe0 |
| the number of the very first frame More... | |
| int | sub_iframe |
| int | mx_empty_word_cnt |
| double | score_offset |
| scaling offset for all scores More... | |
| float | best_WE_score |
| score of the best word final token More... | |
| float | estim_best_score |
| int | estim_best_state |
| the state to be used for updating the estim_best_score field More... | |
| float | best_score |
| score of the best token (so far) More... | |
| float | limit_score |
| best score minus the pruning threshold More... | |
| float | adapt_threshold |
| adaptive version of the pruning threshold, predicted version More... | |
| float | adapt_threshold0 |
| float | adapt_thr_prev |
| float | adapt_log_bw |
| log of the previous beam width More... | |
| float | adapt_log_bw_max |
| log of the maximum beam width More... | |
| float | adapt_thr_bw_stats [4] |
| float | threshold_max |
| maximum pruning threshold (specified by the user) More... | |
| float | threshold_frac |
| maximum increase in threshold to asure a minimum beam_width More... | |
| float | word_threshold_frac |
| float | asf |
| acoustic score scale factor More... | |
| float | H_lim0 |
| float | H_lim1 |
| float | H_factor |
| float | H_av_flt |
| float | H_av |
| float | N_av_tgt |
| maximal beam width over a long interval More... | |
| float | N_av_tgt_flt |
| maximal beam width over a long interval (go there slowly) More... | |
| float | N_av_flt |
| float | N_av |
| average beam width over a long interval More... | |
| float | N_av_ss |
| react differently on sentence starts and silences More... | |
| SprCwrSearchAdaptSpkr * | spkr_settings |
| current speaker specific settings More... | |
| SprCwrSearchAdaptSpkr * | spkr_list |
| list that contains all speaker specific settings (LRU-cache) More... | |
| int * | score_hist |
| int | score_hist_size |
| the size (nr. of tabs) of the score histogram More... | |
| SprCwrSearchResult | topn |
| SprCwrSearchResult | best |
| the best scoring sentence, word sequence or word More... | |
| double | recog_score |
| score assigned to the recognition result More... | |
| SprCwrBeamHandler * | beam_handler |
| SprCwrSearchUpdate * | update_block |
| Request an update of the decoder parameters. More... | |
things needed for this implementation (the standard implementation)
| unsigned int SprCwrSearchStd::load_flags |
which parts (phone_set, lexicon, hmm, ...) are loaded
| unsigned int SprCwrSearchStd::modif_must_do |
combine reset and free operations during modifications
| unsigned int SprCwrSearchStd::modif_have_done |
combine reset and free operations during modifications
| char SprCwrSearchStd::store_seq |
do (not) store the node/phoneme sequences during the search
| char SprCwrSearchStd::recog_mode |
recognition mode
| char SprCwrSearchStd::lm_fwd_mode |
type of LM-forwarding that is used
| char SprCwrSearchStd::sent_end_flags |
for sentence/word end in CWR_best_sentence()
| char SprCwrSearchStd::wexpand_flags |
word expand flags
| char SprCwrSearchStd::sdet_avail |
the preprocessing gives output on sentence boundaries
| char SprCwrSearchStd::sdet_cont |
the recognition stopped at a sentence boundary, further data is available (continuous loop)
| char SprCwrSearchStd::sdet_new_res |
the recognition has found some new non silence frames after a sentence boundary, and has thus produced some new results
| char SprCwrSearchStd::gauss_sel |
gaussian selection status
bit 0: do gaussian selection (1/0); bit 1: gaussian selection must be initialized
| int SprCwrSearchStd::sdet_count |
skip pause between sentences with a periodical return to the main program
| int SprCwrSearchStd::sdet_out |
the preprocessing may mark whether the frame is spch or not (sentence detection)
| SprCwrWordSet SprCwrSearchStd::sentence_begin |
word index/indices of <s>
| SprCwrWordSet SprCwrSearchStd::sentence_end |
word index/indices of </s>
| float SprCwrSearchStd::lm_A |
| float SprCwrSearchStd::lm_C |
upgrade the language model cost: cost = LM_cost*A+C
| unsigned int SprCwrSearchStd::lm_cache_size |
target size of the LM-cache
| unsigned int SprCwrSearchStd::lmc_hash_size |
target size of the LM-context hashing
| float* SprCwrSearchStd::slmsp |
max. unigram probabilities (unscaled, shifted to account for negative indices) trivial way to forward the LM influence
| float* SprCwrSearchStd::slmsp0 |
the unshifted unigram prob array
| int* SprCwrSearchStd::slmsp_wset_mx |
word with best unigram probability in a word set
| int* SprCwrSearchStd::fwd_comb_inst |
definition of word sets when using full LM-forwarding
| int** SprCwrSearchStd::fwd_icomb_inst |
instructions for a specific word set
| int SprCwrSearchStd::fwd_wset_len |
number of new word sets introduced for full LM-forwarding
| int SprCwrSearchStd::Ntokens_max |
maximal beam width (absolute limit on the nr. of open token)
| int SprCwrSearchStd::Ntokens_min |
minimal beam width (absolute limit on the nr. of open token)
| int SprCwrSearchStd::Ntokens_wstart |
number of new tokens created in word end splitting (from group to individual words) and cross word transitions (word startup)
| int SprCwrSearchStd::Ntokens_wend |
number of tokens propagated to a word end position (after observe+jump & pruning)
| int SprCwrSearchStd::Ntokens_wint |
number of tokens propagated to a word internal position (after observe+jump & pruning)
| int SprCwrSearchStd::Nctokens_wend |
number of ctokens propagated to a word end position (after observe+jump & pruning)
| int SprCwrSearchStd::Nctokens_wint |
number of ctokens propagated to a word internal position (after observe+jump & pruning)
| unsigned int SprCwrSearchStd::beam_corr_up |
nr. of correction to the threshold in order to get a larger beam
| unsigned int SprCwrSearchStd::beam_corr_down |
nr. of correction to the threshold in order to get a smaller beam
| unsigned long long SprCwrSearchStd::Ntokens_av_ax |
maximum beam size after word end expansion
| unsigned long long SprCwrSearchStd::Ntokens_av_bp |
maximum beam size before pruning
| unsigned long long SprCwrSearchStd::Ntokens_av_ap |
maximum beam size after pruning
| double SprCwrSearchStd::av_threshold |
average pruning threshold (adaptive)
| unsigned int SprCwrSearchStd::frame_cnt |
nr. of frames used to make up the token scores (also used to calculate some statistics)
| int SprCwrSearchStd::iframe |
number of the current frame
| int SprCwrSearchStd::iframe0 |
the number of the very first frame
| int SprCwrSearchStd::sub_iframe |
sub divide one frame as to have an ordining for the mepty words
| int SprCwrSearchStd::mx_empty_word_cnt |
maximum number of empty words that can be insterted in a recognized string
| double SprCwrSearchStd::score_offset |
scaling offset for all scores
| float SprCwrSearchStd::best_WE_score |
score of the best word final token
| float SprCwrSearchStd::estim_best_score |
score of the best token refering to a non word final lexicon node, used to estimate the best score in the next frame
| int SprCwrSearchStd::estim_best_state |
the state to be used for updating the estim_best_score field
| float SprCwrSearchStd::best_score |
score of the best token (so far)
| float SprCwrSearchStd::limit_score |
best score minus the pruning threshold
| float SprCwrSearchStd::adapt_threshold |
adaptive version of the pruning threshold, predicted version
| float SprCwrSearchStd::adapt_threshold0 |
adaptive version of the pruning threshold, predicted version without acoustic entropy correction
| float SprCwrSearchStd::adapt_thr_prev |
adaptive version of the pruning threshold applied for the previous frame
| float SprCwrSearchStd::adapt_log_bw |
log of the previous beam width
| float SprCwrSearchStd::adapt_log_bw_max |
log of the maximum beam width
| float SprCwrSearchStd::adapt_thr_bw_stats[4] |
learn the relation between y=delta(threshold) and x=f(delta(log(bw)),log(bw),H_av); adapt_thr_bw_stats stores [xy=flt1_w(x*y),xx=flt1_w(x*x),flt2(o=y-xy/xx),o] statistics with flt1_w() a 1st order weighted filter, flt2() a 1st order filter; see beam2threshold_update() for details about f() and the filters flt1() and flt2()
| float SprCwrSearchStd::threshold_max |
maximum pruning threshold (specified by the user)
| float SprCwrSearchStd::threshold_frac |
maximum increase in threshold to asure a minimum beam_width
| float SprCwrSearchStd::word_threshold_frac |
fraction of the threshold used when new words should be started
| float SprCwrSearchStd::asf |
acoustic score scale factor
| float SprCwrSearchStd::H_lim0 |
acoustic entropy based pruning: entropies below this value are OK
| float SprCwrSearchStd::H_lim1 |
acoustic entropy based pruning: entropies above this value get maximal correction of the threshold
| float SprCwrSearchStd::H_factor |
acoustic entropy based pruning: decrease pruning theshold with max(min(H_av_flt,H_lim1)-H_lim0,0)*H_factor
| float SprCwrSearchStd::H_av_flt |
acoustic entropy based pruning: time constant of the first order averaging filter
| float SprCwrSearchStd::H_av |
acoustic entropy based pruning: the average entropy over the last H_av frames (first order filter)
| float SprCwrSearchStd::N_av_tgt |
maximal beam width over a long interval
| float SprCwrSearchStd::N_av_tgt_flt |
maximal beam width over a long interval (go there slowly)
| float SprCwrSearchStd::N_av_flt |
average beam width over a long interval: time constant of the first order averaging filter
| float SprCwrSearchStd::N_av |
average beam width over a long interval
| float SprCwrSearchStd::N_av_ss |
react differently on sentence starts and silences
| SprCwrSearchAdaptSpkr* SprCwrSearchStd::spkr_settings |
current speaker specific settings
| SprCwrSearchAdaptSpkr* SprCwrSearchStd::spkr_list |
list that contains all speaker specific settings (LRU-cache)
| int* SprCwrSearchStd::score_hist |
histogram, used in an algorithme to quickly find the Ntokens_max best hypotheses
| int SprCwrSearchStd::score_hist_size |
the size (nr. of tabs) of the score histogram
| SprCwrSearchResult SprCwrSearchStd::topn |
the best (sorted array) (partial) recognition results in CWR and IWR mode
| SprCwrSearchResult SprCwrSearchStd::best |
the best scoring sentence, word sequence or word
| double SprCwrSearchStd::recog_score |
score assigned to the recognition result
| SprCwrBeamHandler* SprCwrSearchStd::beam_handler |
trap-function to extract information on the content of the search beam at every frame
| SprCwrSearchUpdate* SprCwrSearchStd::update_block |
Request an update of the decoder parameters.
1.8.6
