Mask estimation device, mask estimation method, and mask estimation program
A mask estimation apparatus includes processing circuitry configured to estimate, for a target segment to be processed among a plurality of segments of a continuous time, a first mask which is an occupancy ratio of a target signal to an observation signal of the target segment, based on a first feature obtained from a plurality of the observation signals of the target segment recorded at a plurality of locations, and estimate a parameter for modeling a second feature and a second mask which is an occupancy ratio of the target signal to the observation signal based on an estimation result of the first mask in the target segment and the second feature obtained from the plurality of the observation signals of the target segment.
1. A mask estimation apparatus comprising:
processing circuitry configured to sequentially perform for a plurality of segments of continuous time:
online estimation, for a target segment to be processed among the plurality of segments of continuous time, of a first mask which is an occupancy ratio of a target signal to an observation signal of the target segment, based on a first feature obtained from a plurality of the observation signals of the target segment recorded at a plurality of locations, the occupancy ratio indicating a probability of a target sound occupying each time frequency point of the observation signal; and
online estimation of a parameter for modeling a second feature and a second mask which is an occupancy ratio of the target signal to the observation signal based on an estimation result of the first mask in the target segment and the second feature obtained from the plurality of the observation signals of the target segment, wherein the processing circuitry is further configured to:
update the parameter based on a cumulative sum of a plurality of first masks up to the target segment, and the second feature of the target segment and the second mask,
update the second mask based on the second feature of the target segment, the first mask, and the parameter, and
repeatedly perform the update of the parameter and the second mask until a predetermined convergence condition is satisfied,
wherein the cumulative sum of the plurality of the first masks indicates an amount of target signals included in the observation signal up to the target segment which is a segment for which mask estimation is to be performed,
the cumulative sum of the plurality of first masks indicates a cumulative sum of the probability that a signal of the target sound occupies the observation signal, and
the processing circuitry is further configured to determine whether the cumulative sum of the probability up to the time of the target segment exceeds a predetermined threshold value.
2. The mask estimation apparatus according to claim 1 , wherein the processing circuitry is further configured to:
estimate the first mask using a neural network, and
estimate the second mask based on the first mask and a distribution model of the second feature with the parameter as a condition.
3. The mask estimation apparatus according to claim 1 , wherein the processing circuitry is further configured to:
estimate the second mask based on the estimated parameter in a case where the amount of the plurality of the target signals exceeds the threshold value, and
substitute the estimation result of the first mask for an estimated value of the second mask in a case where the amount of the plurality of the target signals does not exceed the threshold value.
4. A mask estimation method comprising:
sequentially performing for a plurality of segments of continuous time:
online estimation, for a target segment to be processed among the plurality of segments of continuous time, of a first mask which is an occupancy ratio of a target signal to an observation signal of the target segment, based on a first feature obtained from a plurality of the observation signals of the target segment recorded at a plurality of locations, the occupancy ratio indicating a probability of a target sound occupying each time frequency point of the observation signal; and
online estimation of a parameter for modeling a second feature and a second mask which is an occupancy ratio of the target signal to the observation signal based on an estimation result of the first mask in the target segment and the second feature obtained from the plurality of the observation signals of the target segment, by processing circuitry, wherein the method further comprises:
updating the parameter based on a cumulative sum of a plurality of first masks up to the target segment, and the second feature of the target segment and the second mask,
updating the second mask based on the second feature of the target segment, the first mask, and the parameter, and
repeatedly performing the update of the parameter and the second mask until a predetermined convergence condition is satisfied,
wherein the cumulative sum of the plurality of the first masks indicates an amount of target signals included in the observation signal up to the target segment which is a segment for which mask estimation is to be performed,
the cumulative sum of the plurality of first masks indicates a cumulative sum of the probability that a signal of the target sound occupies the observation signal, and
the method further includes determining whether the cumulative sum of the probability up to the time of the target segment exceeds a predetermined threshold value.
5. A non-transitory computer-readable recording medium storing therein a mask estimation program that causes a computer to execute a process comprising:
sequentially performing for a plurality of segments of continuous time:
online estimation, for a target segment to be processed among the plurality of segments of continuous time, of a first mask which is an occupancy ratio of a target signal to an observation signal of the target segment, based on a first feature obtained from a plurality of the observation signals of the target segment recorded at a plurality of locations, the occupancy ratio indicating a probability of a target sound occupying each time frequency point of the observation signal; and
online estimation of a parameter for modeling a second feature and a second mask which is an occupancy ratio of the target signal to the observation signal based on an estimation result of the first mask in the target segment and the second feature obtained from the plurality of the observation signals of the target segment, wherein the process further comprises:
updating the parameter based on a cumulative sum of a plurality of first masks up to the target segment, and the second feature of the target segment and the second mask,
updating the second mask based on the second feature of the target segment, the first mask, and the parameter, and
repeatedly performing the update of the parameter and the second mask until a predetermined convergence condition is satisfied,
wherein the cumulative sum of the plurality of the first masks indicates an amount of target signals included in the observation signal up to the target segment which is a segment for which mask estimation is to be performed,
the cumulative sum of the plurality of first masks indicates a cumulative sum of the probability that a signal of the target sound occupies the observation signal, and
the process further includes determining whether the cumulative sum of the probability up to the time of the target segment exceeds a predetermined threshold value.