Speaker state detecting apparatus and speaker state detecting method

A speaker state detecting method comprising:

• acquiring, at least, a first voice emanated by a first speaker and a second voice emanated by a second speaker by an audio input unit;
• detecting an overlap period between a first speech period of the first speaker included in the first voice and a second speech period of the second speaker included in the second voice, which starts before the first speech period, or an interval between the first speech period and the second speech period by a processor;
• extracting first state information representing a state of the first speaker from the first speech period and second state information representing a state of the second speaker from the second speech period by the processor;
• and
• detecting the state of the first speaker in the first speech period based on the overlap period or the interval and the first and second state information by the processor,
• the detecting the state of the first speaker comprising:
• detecting a state of the second speaker in the second speech period based on the second state information;
• detecting a state of the first speaker in the first speech period based on the first state information;
• deriving a degree of accuracy representing a likelihood of the state of the first speaker;
• and
• determining the detected state of the first speaker to be a state of the first speaker in the first speech period when the degree of accuracy is higher than a redetermination threshold value, on the other hand, when the degree of accuracy is equal to or lower than the redetermination threshold value, obtaining probabilities of possible states which the first speaker can have, corresponding to a set of the overlap period or the interval and the state of the second speaker in the second speech period in accordance with a state affection model, which includes, with respect to a set of an overlap period or interval between two speech periods that are temporally continuous and a state of a speaker who has emanated a voice in a preceding speech period of the two speech periods, probabilities of respective possible states which a speaker who has emanated a voice in a later speech period of the two speech periods can have, and determining a state for which the probability is the maximum, of the possible states which the first speaker can have, to be the state of the first speaker in the first speech period.

A non-transitory computer-readable recording medium in which a computer program for detecting a state of a speaker, the program causing a computer to execute operations, the operations comprising:

• acquiring, at least, a first voice emanated by a first speaker and a second voice emanated by a second speaker;
• detecting an overlap period between a first speech period of the first speaker included in the first voice and a second speech period of the second speaker included in the second voice, which starts before the first speech period, or an interval between the first speech period and the second speech period;
• extracting first state information representing a state of the first speaker from the first speech period and second state information representing a state of the second speaker from the second speech period;
• and
• detecting the state of the first speaker in the first speech period based on the overlap period or the interval and the first and second state information,
• the detecting the state of the first speaker comprising:
• detecting a state of the second speaker in the second speech period based on the second state information;
• detecting a state of the first speaker in the first speech period based on the first state information;
• deriving a degree of accuracy representing a likelihood of the state of the first speaker;
• and
• determining the detected state of the first speaker to be a state of the first speaker in the first speech period when the degree of accuracy is higher than a redetermination threshold value, and when the degree of accuracy is equal to or lower than the redetermination threshold value, obtaining probabilities of possible states which the first speaker can have, corresponding to a set of the overlap period or the interval and the state of the second speaker in the second speech period in accordance with a state affection model, which includes, with respect to a set of an overlap period or interval between two speech periods that are temporally continuous and a state of a speaker who has emanated a voice in a preceding speech period of the two speech periods, probabilities of respective possible states which a speaker who has emanated a voice in a later speech period of the two speech periods can have, and determining a state for which the probability is the maximum, of the possible states which the first speaker can have, to be the state of the first speaker in the first speech period.