# US 5,706,397 A · Speech recognition system with multi-level pruning for acoustic matching > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 5,706,397 A - Document ID: US5706397A - Status: Granted - Kind: A - Application number: 08/539,346 - Filed: 1995-10-05 - Granted: 1998-01-06 - Earliest claimed: 1995-10-05 - Assignees: Apple Computer, Inc. - Inventors: Yen-Lu Chow - Web page: https://patentgrep.com/patent/US5706397A ## CPC - G10L 15/08: Speech classification or search - G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING ## Abstract A method of constructing a new active list of phone models from an existing active list of phone models during acoustic matching of a speech recognition system is described. A vector quantized speech vector is compared against each of the phone models in the existing active list to obtain a phone best score for each of the phone models of the existing active list. A best phone best score is determined among all the phone best scores of the phone models to obtain a global best score. A phone model of the phone models from the existing active list is added to the new active list of phone models if the phone best score of that phone model is within a first predetermined value of the global best score. A next phone model of the existing phone of the existing active list is added to the new active list if the phone ending score of that existing phone is within a second predetermined value of a best score of the existing phone model. A next (e.g. first) phone model of a next word of a particular phone model of the existing active list is added to the new active list if the ending score of that particular phone model is within a third predetermined value of the global best score. ## Claims ### Claim 1 A method of constructing a new active list of phone models from an existing active list of phone models during acoustic matching of a speech recognition system, comprising the steps of: (A) comparing a speech frame against each of the phone models in the existing active list to obtain a phone best score for each of the phone models of the existing active list; (B) adding a next phone model of a first phone model of the phone models of the existing active list to the new active list if the phone best score of the first phone model is within a first predetermined value of a phone ending score of the first phone model. ### Claim 2 (depends on 1) The method of claim 1, further comprising the steps of (a) determining a best phone best score among all the phone best scores of the phone models to obtain a global best score; (b) adding the first phone model to the new active list if the phone best score of the first phone model is within a second predetermined value of the global best score and wherein said first phone model is for a word. ### Claim 3 (depends on 2) The method of claim 2, wherein the phone ending score relative to the phone best score for the first phone model indicates that the acoustic matching should propagate to the next phone model of the first phone model. ### Claim 4 (depends on 3) The method of claim 3, wherein the phone ending score is obtained through training. ### Claim 5 (depends on 4) The method of claim 4, further comprising the step of (C) repeating the steps (B) and (b) for all phone models of the existing active list. ### Claim 6 (depends on 5) The method of claim 5, further comprising the steps of (i) determining if a new speech frame has been received; (ii) repeating the steps (A) through (C) using the new active list of phone models as the existing active list if the new speech frame has been received. ### Claim 7 A method of constructing a new active list of phone models from an existing active list of phone models during acoustic matching of a speech recognition system, comprising the steps of: (A) comparing a speech frame against each of the phone models in the existing active list to obtain a phone best score for each of the phone models of the existing active list; (B) determining a best phone best score among all the phone best scores of the phone models of the existing active list to obtain a global best score; (C) adding a next phone model of a next word of a first phone model of the phone models of the existing active list to the new active list if the global best score is within a first predetermined value with respect to a word ending score of the first phone model. ### Claim 8 (depends on 7) The method of claim 7, further comprising the step of (D) repeating the step (C) for all phone models of the existing active list. ### Claim 9 (depends on 8) The method of claim 8, further comprising the steps of (i) determining if a new speech frame has been received; (ii) repeating the steps (A) through (D) if the new speech frame has been received. ### Claim 10 (depends on 9) The method of claim 9, wherein the word ending score for the first phone model indicates that the acoustic matching should propagate to the next phone model of the next word of the first phone model, wherein the next phone model of the next word of the first phone model is determined by a predetermined language model. ### Claim 11 (depends on 10) The method of claim 10, wherein the word ending score is obtained through the language model. ### Claim 12 (depends on 7) The method of claim 7, further comprising the step of adding the first phone model to the new active list if the phone best score of the first phone model is within a second predetermined value of the global best score. ### Claim 13 A method of constructing a new active list of phone models from an existing active list of phone models during acoustic matching of a speech recognition system, comprising the steps of: (A) comparing a speech frame against each of the phone models of the existing active list to obtain a phone best score for each of the phone models of the existing active list; (B) determining a best phone best score among all the phone best scores of the phone models from the existing active list to obtain a global best score; (C) adding a first phone model of the phone models from the existing active list to the new active list of phone models if the phone best score of the first phone model is within a first predetermined value of the global best score; (D) adding a next phone model of the first phone model to the new active list if the phone best score of the first phone model is within a second predetermined value of a word ending score of the first phone model; (E) adding a next phone model of a next word of the first phone model to the new active list if the global best score is within a third predetermined value with respect to a word ending score of the first phone model. ### Claim 14 (depends on 13) The method of claim 13, wherein the predetermined phone ending score for the first phone model indicates that the acoustic matching should propagate to the next phone model of the first phone model. ### Claim 15 (depends on 14) The method of claim 14, wherein the second predetermined value is obtained through training. ### Claim 16 (depends on 13) The method of claim 13, wherein the word ending score for the first phone model indicates that the acoustic matching should propagate to the next phone model of the next word, wherein the next phone model of the next word is determined by a predetermined language model. ### Claim 17 (depends on 16) The method of claim 16, wherein the second predetermined value is obtained through training. ### Claim 18 (depends on 13) The method of claim 13, further comprising the step of (F) repeating the steps (C) through (E) for all phone models of the existing active list. ### Claim 19 (depends on 18) The method of claim 18, further comprising the step of (i) determining if a new speech frame has been received; (ii) repeating the steps (A) through (F) using the new active list of phone models as the existing active list if the new speech frame has been received. ### Claim 20 A speech recognition apparatus comprising an audio transducer which receives a speech utterance; a processor coupled to said audio transducer to receive a speech frame of said speech utterance; a memory coupled to said processor, said memory storing a plurality of phone models in an existing active list of phone models, said processor pruning said existing active list to create a new active list of phone models, said processor comparing said speech frame against each of the phone models in the existing active list to obtain a phone best score for each of the phone models of the existing active list, and adding a next phone model of a first phone model of the phone models of the existing active list to the new active list if the phone best score of the first phone model is within a first predetermined value of a phone ending score of the first phone model. ### Claim 21 A speech recognition apparatus comprising an audio transducer which receives a speech utterance; a processor coupled to said audio transducer to receive a speech frame of said speech utterance; a memory coupled to said processor, said memory storing a plurality of phone models in an existing active list of phone models, said processor pruning said existing active list to create a new active list, said processor comparing said speech frame against each of the phone models in the existing active list to obtain a phone best score for each of the phone models of the existing active list; determining a best phone best score among all the phone best scores of the phone models of the existing active list to obtain a global best score; and adding a next phone model of a next word of a first phone model of the phone models of the existing active list to the new active list if the global best score is within a first predetermined value with respect to a word ending score of the first phone model.