Speech recognition method, speech recognition apparatus, and non-transitory computer-readable recording medium storing a program

A speech recognition apparatus that recognizes speech from sound information acquired by a plurality of microphones, comprising:

• a sound information acquirer that acquires sound information from the plurality of microphones;
• a sound source interval detector that detects a sound source interval including sound from the sound information;
• a direction estimator that acquires by direction estimation an estimated direction of a speech interval from among the sound source interval;
• an adaptation processor that conducts an adaptation process of using the sound information to estimate filter coefficients for extracting speech in the estimated direction;
• an adaptation process convergence monitor that acquires information about a convergence state of the adaptation process;
• a buffer that holds the sound information according to a decided buffer size;
• a buffer size decider that decides the buffer size of the sound information to hold in the buffer, based on information about the sound source interval, information about the estimated direction, and information about the convergence state of the adaptation process;
• a beamforming processor that conducts a beamforming process using the sound information held in the buffer and the filter coefficients, and acquires speech information;
• and
• a speech recognizer that conducts speech recognition on the speech information acquired by the beamforming process, wherein
• immediately after processing of the sound information starts, the buffer size decider decides a buffer size of a size sufficient for the convergence of the adaptation process as the buffer size to hold in the buffer.

A non-transitory computer-readable recording medium storing a program, the program causing a computer to execute processing comprising:

• (c1) acquiring sound information from a plurality of microphones;
• (c2) detecting a sound source interval including sound from the sound information;
• (c3) acquiring by direction estimation an estimated direction of a speech interval from among the sound source interval;
• (c4) estimating, by an adaptation process using the sound information, filter coefficients for extracting speech information in the estimated direction;
• (c5) deciding a buffer size of the sound information to hold in a buffer, based on information about the sound source interval, information about the estimated direction, and information about a convergence state of the adaptation process;
• (c6) holding the sound information in the buffer according to the decided buffer size;
• (c7) conducting a beamforming process using the sound information held in the buffer and the filter coefficients, and acquiring speech information;
• and
• (c8) conducting speech recognition on the speech information acquired by the beamforming process, wherein
• immediately after processing of the sound information starts, the sound information of a size sufficient for the convergence of the adaptation process is held in the buffer.