# US 10,373,607 B2 · Testing words in a pronunciation lexicon > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 10,373,607 B2 - Document ID: US10373607B2 - Status: Granted - Kind: B2 - Application number: 15/621,778 - Filed: 2017-06-13 - Published: 2017-09-28 - Granted: 2019-08-06 - Earliest claimed: 2015-06-30 - Assignees: International Business Machines Corporation - Applicants: International Business Machines Corporation - Inventors: Takashi Fukuda; Osamu Ichikawa; Futoshi Iwama - Web page: https://patentgrep.com/patent/US10373607B2 ## CPC - G10L 15/01: Assessment or evaluation of speech recognition systems - G10L 13/08: Text analysis or generation of parameters for speech synthesis out of text, e.g. grapheme to phoneme translation, prosody generation or stress or intonation determination - G10L 15/193: Formal grammars, e.g. finite state automata, context free grammars or word networks - G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING ## Abstract A method, for testing words defined in a pronunciation lexicon used in an automatic speech recognition (ASR) system, is provided. The method includes: obtaining test sentences which can be accepted by a language model used in the ASR system. The test sentences cover words defined in the pronunciation lexicon. The method further includes obtaining variations of speech data corresponding to each test sentence, and obtaining a plurality of texts by recognizing the variations of speech data, or a plurality of texts generated by recognizing the variation of speech data. The method also includes constructing a word graph, using the plurality of texts, for each test sentence, where each word in the word graph corresponds to each word defined in the pronunciation lexicon; and determining whether or not all or parts of words in a test sentence are present in a path of the word graph derived from the test sentence. ## Claims ### Claim 1 A computer-implemented method for testing words defined in a pronunciation lexicon used in an automatic speech recognition system, comprising: obtaining a plurality of test sentences which can be accepted by a language model used in the automatic speech recognition system, the test sentences including a plurality of the words defined in the pronunciation lexicon; obtaining variations of speech data corresponding to each of the test sentences; obtaining a plurality of texts generated by recognizing the variations of speech data; constructing a word graph, using the plurality of texts, for each of the test sentences, wherein one or more words in the word graph correspond to one or more words defined in the pronunciation lexicon; and determining whether all, or parts of words in a test sentence of the test sentences are present in a path of the word graph derived from the test sentence. ### Claim 2 (depends on 1) The method according to claim 1, wherein the test sentences cover all of partial structures of the language model. ### Claim 3 (depends on 2) The method according to claim 2, wherein the partial structures of the language model are edges in Finite State Transducer, or production rules and alternatives of each rule in Context Free Grammar. ### Claim 4 (depends on 3) The method according to claim 3, wherein each alternative corresponds to a right-hand side choice of the production rules. ### Claim 5 (depends on 1) The method according to claim 1, wherein each of the test sentences is generated together with feature information obtained from the language model and wherein a number of the variations of speech data is determined according to corresponding feature information. ### Claim 6 (depends on 5) The method according to claim 5, wherein the feature information comprises a number of transitions to other words in the language model, a transition probability among words in the language model, or a combination thereof. ### Claim 7 (depends on 1) The method according to claim 1, further comprising, in a case where all of the words in a test sentence of the test sentences are present in a path of the word graph derived from the test sentence, judging whether all of the words in the test sentence are correctly accepted in the automatic speech recognition system. ### Claim 8 (depends on 1) The method according to claim 1, further comprising, in a case where parts of words in a test sentence of the test sentences are present in a path of the word graph derived from the test sentence, judging whether matching parts of the words in the test sentence are correctly accepted in the automatic speech recognition system. ### Claim 9 A computer system for testing words defined in a pronunciation lexicon, comprising: an automatic speech recognition system (ASR) for using the pronunciation lexicon, the ASR system configured to generate a plurality of test sentences which can be accepted by a language model used in the ASR system, the test sentences including a plurality of the words defined in the pronunciation lexicon; and a text-to-speech system configured to generate variations of speech data corresponding to each of the test sentences; wherein the automatic speech recognition system is further configured to: obtain a plurality of texts generated by recognizing the variations of speech data, construct a word graph, using the plurality of texts, for each of the test sentences, wherein one or more words in the word graph correspond to one or more words defined in the pronunciation lexicon, and determine whether all, or parts of words in a test sentence of the test sentences are present in a path of the word graph derived from the test sentence. ### Claim 10 (depends on 9) The computer system according to claim 9, wherein the test sentences cover all of partial structures of the language model. ### Claim 11 (depends on 10) The computer system according to claim 10, wherein the partial structures of the language model are edges in Finite State Transducer, or production rules and alternatives of each rule in Context Free Grammar. ### Claim 12 (depends on 9) The computer system according to claim 9, wherein each of the test sentences is generated together with feature information obtained from the language model and a number of the variations of speech data is determined according to corresponding feature information. ### Claim 13 (depends on 12) The computer system according to claim 12, wherein the feature information comprises a number of transitions to other words in the language model, a transition probability among words in the language model, or a combination thereof. ### Claim 14 (depends on 9) The computer system according to claim 9, in a case where all of the words in a test sentence of the test sentences are present in a path of the word graph derived from the test sentence, the system further judges that all of the words in the test sentence are correctly accepted in the automatic speech recognition system. ### Claim 15 A computer program product comprising a computer readable storage medium including computer usable code implemented in one more of computers, for testing words defined in a pronunciation lexicon used in an automatic speech recognition system, comprising: computer usable code for: obtaining a plurality of test sentences which can be accepted by a language model used in the automatic speech recognition system, the test sentences including a plurality of the words defined in the pronunciation lexicon; obtaining variations of speech data corresponding to each of the test sentences; obtaining a plurality of texts generated by recognizing the variations of speech data; constructing a word graph, using the plurality of texts, for each of the test sentences, wherein one or more words in the word graph correspond to one or more words defined in the pronunciation lexicon; and determining whether all, or parts of words in a test sentence of the test sentences are present in a path of the word graph derived from the test sentence. ### Claim 16 (depends on 15) The computer program product according to claim 15, wherein the test sentences cover all of partial structures of the language model. ### Claim 17 (depends on 16) The computer program product according to claim 16, wherein the partial structures of the language model are edges in Finite State Transducer, or production rules and alternatives of each rule in Context Free Grammar. ### Claim 18 (depends on 15) The computer program product according to claim 15, wherein each of the test sentences is generated together with feature information obtained from the language model and a number of the variations of speech data is determined according to corresponding feature information. ### Claim 19 (depends on 18) The computer program product according to claim 18, wherein the feature information comprises a number of transitions to other words in the language model, a transition probability among words in the language model, or a combination thereof. ### Claim 20 (depends on 15) The computer program product according to claim 15, further comprising computer usable code for, in a case where all of the words in a test sentence of the test sentences are present in a path of the word graph derived from the test sentence, judging whether all of the words in the test sentence are correctly accepted in the automatic speech recognition system.