Sound processing apparatus, speech recognition apparatus, sound processing method, speech recognition method, storage medium

A speech recognition apparatus comprising:

• a plurality of first microphones that are arranged at positions different from each other in a space comprising an objective area and a place outside the objective area, and obtains a first sound comprising first speech uttered at the place outside the objective area;
• a second microphone that is arranged in the space, and obtains second speech;
• a feature extractor that extracts a first feature from the first sound obtained by the plurality of first microphones, and extracts a second feature from the second speech obtained by the second microphone, the first feature comprising a speech feature corresponding to the first speech obtained by the plurality of first microphones, and a background sound corresponding to sound in the first sound other than the first speech;
• a speech feature extractor configured to extract the speech feature from the first feature, and store the speech feature in a storage;
• a determination model creator that creates a determination model configured to determine whether an utterance position of a portion of the second speech is outside the objective area by learning based at least in part on the speech feature stored in the storage;
• a speech detector that eliminates the portion of the second speech from the second speech obtained by the second microphone when it is determined that the portion of the second speech is uttered at the place outside the objective area based at least in part on the speech feature of the first speech and the determination model, wherein the speech detector detects and outputs remaining speech from the second speech;
• a recognition device that recognizes contents of the remaining speech detected by the speech detector;
• and
• a switch configured to switch between an initial setting mode and an operation mode in response to instructions sent from a controller, wherein the first speech is obtained in the initial setting mode, the second speech is obtained in the operation mode, and the controller compares a feature of the portion of the second speech and the determination model in the operation mode.

A method for sound processing, the method comprising:

• arranging a plurality of first microphones at positions different from each other in a space comprising an objective are and a place outside the objective area;
• obtaining a first sound comprising first speech uttered at the place outside the objective area by the plurality of first microphones;
• arranging a second microphone in the space;
• obtaining second speech by the second microphone;
• extracting a first feature from the first sound obtained by the plurality of first microphones, the first feature comprising a speech feature corresponding to the first speech obtained by the plurality of first microphones, and a background sound corresponding to sound in the first sound other than the first speech;
• extracting a second feature from the second speech;
• extracting the speech feature from the first feature;
• storing the speech feature in a storage;
• creating a determination model configured to determine whether an utterance position of a portion of the second speech is outside the objective area by learning based at least in part on the speech feature stored in the storage;
• and
• eliminating the portion of the second speech from the second speech obtained by the second microphone when it is determined that the portion of the second speech is uttered at the place outside the objective area based at least in part on the speech feature of the first speech and the determination model;
• detecting remaining speech from the second speech;
• outputting the remaining speech;
• switching between an initial setting mode and an operation mode in response to instructions sent from a controller, wherein the first speech is obtained in the initial setting mode, the second speech is obtained in the operation mode;
• and
• comparing a feature of the portion of the second speech and the determination model in the operation mode.

A method of speech recognition, the method comprising:

• arranging a plurality of first microphones at positions different from each other in a space comprising an objective area and a place outside an objective area;
• obtaining a first sound comprising first speech uttered at the place outside the objective area by the plurality of first microphones;
• arranging a second microphone in the space;
• obtaining second speech by the second microphone;
• extracting a first feature from first sound obtained by the plurality of first microphones, the first feature comprising a speech feature corresponding to the first speech obtained by the plurality of first microphones, and a background sound corresponding to sound in the first sound other than the first speech;
• extracting a second feature from the second speech;
• extracting the speech feature from the first feature;
• storing the speech feature in a storage;
• creating a determination model configured to determine whether an utterance position of a portion of the second speech is outside the objective area by learning based at least in part on the speech feature stored in the storage;
• eliminating the portion of the second speech from the second speech obtained by the second microphone when it is determined that the portion of the second speech is uttered at the place outside the objective area based at least in part on the speech feature of the first speech and the determination model;
• detecting remaining speech from the second speech;
• recognizing contents of the remaining speech;
• switching between an initial setting mode and an operation mode in response to instructions sent from a controller, wherein the first speech is obtained in the initial setting mode, the second speech is obtained in the operation mode;
• and
• comparing a feature of the portion of the second speech and the determination model in the operation mode.

A non-transitory computer-readable storage comprising a computer program that is executable by a computer used in a sound processing program, the computer program comprising instructions for causing the computer to execute functions of:

• extracting a first feature from a first sound comprising first speech uttered at a place outside an objective area, and obtained by a plurality of first microphones that are arranged at positions different from each other in a space comprising the objective area and the place outside the objective area, the first feature comprising a speech feature corresponding to the first speech obtained by the plurality of first microphones, and a background sound corresponding to sound in the first sound other than the first speech;
• extracting the speech feature from the first feature;
• storing the speech feature in a storage;
• creating a determination model configured to determine whether an utterance position of a portion of second speech obtained by a second microphone that is arranged in the space is outside the objective area by learning based at least in part on the speech feature stored in the storage;
• and
• eliminating the portion of the second speech from the second speech when it is determined that the portion of the second speech is uttered at the place outside the objective area based at least in part on the speech feature of the first speech and the determination model;
• detecting remaining speech from the second speech;
• outputting the remaining speech;
• switching between an initial setting mode and an operation mode in response to instructions sent from a controller, wherein the first speech is obtained in the initial setting mode, the second speech is obtained in the operation mode;
• and
• compares a feature of the portion of the second speech and the determination model in the operation mode.

A non-transitory computer-readable storage comprising a computer program that is executable by a computer used in a speech recognition program, the computer program comprising instructions for causing the computer to execute functions of:

• extracting a first feature from a first sound comprising first speech uttered at a place outside an objective area, and obtained by a plurality of first microphones that are arranged at positions different from each other in a space comprising the objective area the place outside the objective area, the first feature comprising a speech feature corresponding to the first speech obtained by the plurality of first microphones, and a background sound corresponding to sound in the first sound other than the first speech;
• extracting the speech feature from the first feature;
• storing the speech feature in a storage;
• creating a determination model configured to determine whether an utterance position of a portion of second speech obtained by a second microphone that is arranged in the space is outside the objective area by learning based at least in part on the speech feature amount of the first speech uttered outside the objective area stored in the storage;
• and
• eliminating the portions of the second speech from the second speech when it is determined that the portion of the second speech is uttered at the place outside the objective area based at least in part on the speech feature of the first speech and the determination model;
• detecting remaining speech from the second speech;
• outputting the remaining speech;
• recognizing contents of the remaining speech;
• switching between an initial setting mode and an operation mode in response to instructions sent from a controller, wherein the first speech is obtained in the initial setting mode, the second speech is obtained in the operation mode;
• and
• compares a feature of the portion of the second speech and the determination model in the operation mode.