# US 2021/0280168 A1 · Speech recognition error correction apparatus > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2021/0280168 A1 - Document ID: US20210280168A1 - Status: Published application - Kind: A1 - Application number: 17/012,372 - Filed: 2020-09-04 - Published: 2021-09-09 - Earliest claimed: 2020-03-04 - Assignees: Kabushiki Kaisha Toshiba - Applicants: Kabushiki Kaisha Toshiba - Inventors: Taira ASHIKAWA; Hiroshi FUJIMURA; Kenji IWATA - Web page: https://patentgrep.com/patent/US20210280168A1 ## CPC - G10L 15/01: Assessment or evaluation of speech recognition systems - G10L 15/193: Formal grammars, e.g. finite state automata, context free grammars or word networks - G10L 15/22: Procedures used during a speech recognition process, e.g. man-machine dialogue - G10L 15/285: {Memory allocation or algorithm optimisation to reduce hardware requirements} - G10L 15/30: Distributed recognition, e.g. in client-server systems, for mobile phones or network applications - G10L 2015/223: {Execution procedure of a spoken command} - G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING ## Abstract According to one embodiment, a speech recognition error correction apparatus includes a correction network memory and an error correction circuitry. The error correction circuitry calculates a difference between a speech recognition result string of an error correction target, which is a result of performing speech recognition on a new series of speech data, and a correction network, where a speech recognition result string and a correction result by a user for the speech recognition result string are associated, and when a value indicating the difference is equal to or less than a threshold, perform error correction on a speech recognition error portion in the speech recognition result string of the error correction target by using the correction network to generate a speech recognition error correction result string. ## Claims ### Claim 1 A speech recognition error correction apparatus comprising: a correction network memory configured to store a correction network, where a speech recognition result string, which is a result of performing speech recognition on a series of speech data, and a correction result by a user for the speech recognition result string are associated; and an error correction circuitry configured to calculate a difference between a speech recognition result string of an error correction target, which is a result of performing speech recognition on a new series of speech data, and the correction network stored in the correction network memory, and when a value indicating the difference is equal to or less than a threshold, perform error correction on a speech recognition error portion in the speech recognition result string of the error correction target by using the correction network to generate a speech recognition error correction result string. ### Claim 2 (depends on 1) The speech recognition error correction apparatus according to claim 1, further comprising: a correction sentence acquisition circuitry configured to acquire the correction result by the user for the speech recognition error correction result string generated by the error correction circuitry; and a generation circuitry configured to calculate a difference between the speech recognition result string of the error correction target and the correction result, and when a value indicating the calculated difference is equal to or less than a threshold, generate the correction network, where the speech recognition result string of the error correction target and the correction result are associated, and store the generated correction network in the correction network memory. ### Claim 3 (depends on 1) The speech recognition error correction apparatus according to claim 1, further comprising a presentation circuitry configured to present the speech recognition error correction result string generated by the error correction circuitry. ### Claim 4 (depends on 1) The speech recognition error correction apparatus according to claim 1, wherein the error correction circuitry is configured to: when the value indicating the difference between the speech recognition result string of the error correction target and the correction network is equal to or less than the threshold, perform the error correction using the correction network; and when the value indicating the difference is larger than the threshold, not perform the error correction using the correction network, and the speech recognition error correction apparatus further comprises a presentation circuitry configured to present the speech recognition error correction result string when the error correction circuitry performs the error correction, and present the speech recognition result string of the error correction target when the error correction circuitry does not perform the error correction. ### Claim 5 (depends on 1) The speech recognition error correction apparatus according to claim 1, further comprising: a conversion rule memory configured to store a predetermined conversion rule, which is a correction rule that commonly applies to a plurality of users; and a dialogue processing circuitry configured to apply the conversion rule stored in the conversion rule memory to the speech recognition error correction result string generated by the error correction circuitry, and correct the speech recognition error correction result string. ### Claim 6 (depends on 5) The speech recognition error correction apparatus according to claim 5, wherein the dialogue processing circuitry presents a correction portion that is corrected by applying the conversion rule in the speech recognition error correction result string. ### Claim 7 (depends on 6) The speech recognition error correction apparatus according to claim 6, wherein the dialogue processing circuitry is configured to receive designation by a user of a portion to be registered in the conversion rule in the speech recognition error correction result string, and the speech recognition error correction apparatus further comprises a presentation circuitry configured to present the speech recognition error correction result string together with the correction portion and the designated designation portion by the dialogue processing circuitry. ### Claim 8 (depends on 7) The speech recognition error correction apparatus according to claim 7, further comprising: a correction sentence acquisition circuitry configured to acquire the correction result by the user for the speech recognition error correction result string generated by the error correction circuitry; and a generation circuitry configured to: extract a correction portion due to a speech recognition error and a correction portion not due to a speech recognition error in the correction result based on the correction portion and the designation portion by the dialogue processing circuitry; for the correction portion due to the speech recognition error, when the value indicating a difference between the speech recognition result string of the error correction target and the correction result is equal to or less than a threshold, generate the correction network, where the speech recognition result string of the error correction target and the correction result are associated, and store the generated correction network in the correction network memory; and for the correction portion not due to the speech recognition error, generate the conversion rule by the difference between the speech recognition result string of the error correction target and the correction result, and store the generated conversion rule in the conversion rule memory. ### Claim 9 (depends on 8) The speech recognition error correction apparatus according to claim 8, further comprising a synthesis circuitry configured to synthesize the conversion rule stored in the conversion rule memory to the correction network stored in the correction network memory. ### Claim 10 (depends on 1) The speech recognition error correction apparatus according to claim 1, wherein the speech recognition result string, the correction result, and the speech recognition error correction result string are one of character strings and word strings. ### Claim 11 (depends on 1) The speech recognition error correction apparatus according to claim 1, wherein the speech recognition result string and the correction network are expressed by at least one of a text, a lattice, which is a candidate network, and a confusion network. ### Claim 12 (depends on 1) The speech recognition error correction apparatus according to claim 1, further comprising: a speech acquisition circuitry configured to acquire the new series of speech data; and a recognition circuitry configured to perform speech recognition on the new series of speech data acquired by the speech acquisition circuitry, and generate the speech recognition result string of the error correction target. ### Claim 13 (depends on 2) The speech recognition error correction apparatus according to claim 2, wherein the difference between the speech recognition result string of the error correction target and the correction network is a shortest path between the speech recognition result string of the error correction target and the correction network, a difference between the speech recognition result string of the error correction target and the correction result is a shortest path between the speech recognition result string of the error correction target and the correction result, and a value indicating the difference is a shortest path score. ### Claim 14 A method for correcting a speech recognition error comprising: storing a correction network, where a speech recognition result string, which is a result of performing speech recognition on a series of speech data, and a correction result by a user for the speech recognition result string are associated, in a correction network memory; calculating a difference between a speech recognition result string of an error correction target, which is a result of performing speech recognition on a new series of speech data, and the correction network stored in the correction network memory; and when a value indicating the difference is equal to or less than a threshold, performing error correction for a speech recognition error portion in the speech recognition result string of the error correction target by using the correction network to generate a speech recognition error correction result string. ### Claim 15 A non-transitory storage medium storing a speech recognition error correction program for causing a computer to: in order to correct a speech recognition error portion in a speech recognition result string, which is a result of performing speech recognition on a series of speech data, perform speech recognition for a new series of speech data to generate a speech recognition result string of an error correction target; calculate a difference between a pre-stored correction network, where the speech recognition result string, which is a result of performing the speech recognition on the series of speech data, and a correction result by a user for the speech recognition result string are associated, and the speech recognition result string of the error correction target; and when a value indicating the difference is equal to or less than a threshold, perform error correction on a speech recognition error portion in the speech recognition result string of the error correction target by using the correction network to generate a speech recognition error correction result string.