SPEAKER PREDICTION METHOD, SPEAKER PREDICTION DEVICE, AND COMMUNICATION SYSTEM

The method and device enhance speaker prediction accuracy by analyzing voice and image features to anticipate the next speaker in conversations, improving communication systems.

JP7577960B2Active Publication Date: 2024-11-06YAMAHA CORP
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
JP2020171050
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Filing Date
2020-10-09
Publication Date
2024-11-06
Estimated Expiration
2040-10-09

AI Technical Summary

Technical Problem

Existing speaker prediction technologies struggle to accurately predict the next speaker during ongoing conversations.

Method used

A method and device that acquires and records voice and conversation history, identifies speakers using voice and image features, and predicts the next speaker based on conversation patterns and probabilities.

Benefits of technology

Enables high-accuracy prediction of the next speaker during ongoing conversations, enhancing communication systems with improved speaker management.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007577960000001
    Figure 0007577960000001
  • Figure 0007577960000002
    Figure 0007577960000002
  • Figure 0007577960000003
    Figure 0007577960000003
Patent Text Reader

Abstract

To provide a speaker prediction method, a speaker prediction device, and a communication system capable of predicting a next speaker with high accuracy while a current speaker is speaking.SOLUTION: A voice processing device includes: a voice acquisition unit 100 for receiving a far-end side sound signal; an image acquisition unit 101 for receiving a far-end side speaker image; a far-end speaker identification unit 104 for identifying a far-end side speaker based on the far-end side sound signal and the far-end side speaker image; a conversation history recording unit 103 for recording an identifier pointing to the identified speaker and the order thereof as a conversation history; and a prediction unit 51 for predicting a next speaker based on the identified speaker and the conversation history.SELECTED DRAWING: Figure 4
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] An embodiment of the present invention relates to a speaker prediction method, a speaker prediction device, and a communication system for predicting a speaker. [Background technology]

[0002] Patent Document 1 discloses a speaker prediction device that predicts who will be the next speaker from among a plurality of people. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2012-29209 A Summary of the Invention [Problem to be solved by the invention]

[0004] It is desirable to predict with greater accuracy who will be the next speaker while the current speaker is speaking.

[0005] Therefore, an object of one embodiment of the present invention is to provide a speaker prediction method, a speaker prediction device, and a communication system that can predict the next speaker to speak with high accuracy while the current speaker is speaking. [Means for solving the problem]

[0006] The speaker prediction method acquires voices of multiple speakers and records the conversation history of the multiple speakers, acquires the voices, identifies the speaker of the acquired voices, and predicts the next speaker to speak from among the multiple speakers based on the identified speaker and the conversation history. Effect of the Invention

[0007] According to one embodiment of the present invention, the next speaker can be predicted with high accuracy while the current speaker is still speaking. [Brief description of the drawings]

[0008] [Figure 1] 1 is a block diagram showing a configuration of a sound signal processing device 1. FIG. [Diagram 2] 1 is a functional block diagram of a sound signal processing device 1. FIG. [Diagram 3] 1 is a functional block diagram of a sound signal processing device 1. FIG. [Figure 4] 1 is a functional block diagram of a sound signal processing device 1. FIG. [Diagram 5] 4 is a flowchart showing the operation of a sound signal processing method. [Figure 6] 4 is a flowchart showing the operation of a sound signal processing method. [Figure 7] 4 is a flowchart showing the operation of a sound signal processing method. [Figure 8] FIG. 2 is a diagram showing an example of an image captured by a camera 11. [Figure 9] FIG. 13 is a diagram showing an example of a conversation history. [Figure 10] 1 is a block diagram showing a functional configuration of a sound signal processing device 1 when performing sound signal processing. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] 1 is a block diagram showing the configuration of a sound signal processing device 1. The sound signal processing device 1 includes a camera 11, a CPU 12, a DSP 13, a flash memory 14, a RAM 15, a user interface (I / F) 16, a speaker 17, six microphones 18A to 18F, and a communication unit 19. In this embodiment, the term "signal" refers to a digital signal.

[0010] The camera 11, the speaker 17, and the microphones 18A to 18F are arranged, for example, above or below a display (not shown). The camera 11 acquires an image of a user in front of the display (not shown). The microphones 18A to 18F acquire the voice of the user in front of the display (not shown). The speaker 17 outputs voice to the user in front of the display (not shown). The number of microphones is not limited to six. The microphone may be one microphone. In this embodiment, the number of microphones is six, forming an array microphone. The DSP 13 performs beamforming processing on the sound signals acquired by the microphones 18A to 18F.

[0011] The CPU 12 functions as a control unit that comprehensively controls the operation of the sound signal processing device 1 by reading out an operating program from the flash memory 14 to the RAM 15. Note that the program does not need to be stored in the flash memory 14 of the device itself. The CPU 12 may download the program from a server or the like and read it into the RAM 15 each time.

[0012] The DSP 13 is a signal processing unit that processes the video signal and the audio signal under the control of the CPU 12. The DSP 13 functions as an image processing unit that performs framing processing to cut out an image of a speaker from a video signal, for example. The DSP 13 also functions as an audio signal processing unit that performs beamforming, for example.

[0013] The communication unit 19 transmits the near-end video signal and audio signal processed by the DSP 13 to another device. The other device is a far-end audio signal processing device connected via the Internet or the like. The communication unit 19 also receives video signals and audio signals from the other device. The communication unit 19 outputs the received video signal to a display (not shown). The communication unit 19 outputs the received audio signal to the speaker 17. The display displays an image captured by a camera of the other device. The speaker 17 outputs the voice of the speaker captured by a microphone of the other device. In this way, the audio signal processing device 1 functions as a communication system for conducting audio conversations with remote locations.

[0014] Figures 2, 3, and 4 are functional block diagrams of the sound signal processing device 1. Figures 5, 6, and 7 are flowcharts showing the operation of the sound signal processing method. The functional configurations shown in Figures 2, 3, and 4 are realized by the CPU 12 and the DSP 13.

[0015] The sound signal processing device 1 functionally comprises a voice acquisition unit 100, an image acquisition unit 101, a near-end speaker identification unit 102, a conversation history recording unit 103, a far-end speaker identification unit 104, a camera image control unit 105, and a prediction unit 51.

[0016] The near-end speaker identification unit 102 and the far-end speaker identification unit 104 correspond to the speaker identification unit of the present invention. Fig. 2 shows a configuration in which speaker identification is performed based on the near-end video signal and audio signal and recorded as a conversation history, and Fig. 5 shows the operation in that case.

[0017] The voice acquiring unit 100 inputs sound signals from the microphones 18A to 18F as near-end sound signals (S11). The image acquiring unit 101 acquires a video signal from the camera 11 as a near-end speaker image (S12). The near-end speaker identifying unit 102 identifies the near-end speaker based on the near-end sound signal and the near-end speaker image (S13). The conversation history recording unit 103 records identifiers indicating the speakers identified by the near-end speaker identifying unit 102 and the order of their speech as a conversation history (S14).

[0018] The near-end speaker identification unit 102 identifies a speaker based on, for example, a voice feature. The voice feature is unique information that differs for each speaker. The voice feature includes, for example, an average volume (power), a maximum pitch frequency, a minimum pitch frequency, or a speech rate. The near-end speaker identification unit 102 extracts voice features from a sound signal and identifies a speaker by a predetermined algorithm using, for example, a neural network. The near-end speaker identification unit 102 may also identify a speaker based on the arrival direction of the voice. For example, the near-end speaker identification unit 102 can determine the timing at which the voice of the speaker reaches the microphone by calculating the cross-correlation of sound signals acquired by multiple microphones. The near-end speaker identification unit 102 can determine the arrival direction of the voice of the speaker based on the positional relationship between the microphones and the arrival timing of the voice.

[0019] The near-end speaker identification unit 102 may also identify a speaker by performing face recognition processing of the speaker. The face recognition processing of the speaker is processing for recognizing the position of the speaker's face from the near-end speaker image by using a predetermined algorithm using, for example, a neural network or the like.

[0020] Fig. 8 is a diagram showing an example of an image captured by the camera 11. In the example of Fig. 8, the camera 11 captures facial images of a plurality of users positioned along the longitudinal direction (depth direction) of the desk T.

[0021] The desk T has a rectangular shape in a plan view. The camera 11 captures images of four users on the left and right sides of the desk T in the short side direction, as well as a user located farther away from the desk T.

[0022] The near-end speaker identification unit 102 recognizes the face of the speaker from such an image captured by the camera 11. In the example of FIG. 8, user A1 located at the bottom left of the image is speaking. Based on multiple frames of images, the near-end speaker identification unit 102 recognizes, for example, a face image with a change in the mouth image as the speaker's face. Note that in the example of FIG. 8, the faces of the other users A2 to A5 are recognized, but they are not speakers. Therefore, the near-end speaker identification unit 102 recognizes the face of user A1 as the speaker's face.

[0023] Further, the near-end speaker identification unit 102 may set a bounding box, as shown by a square in the figure, at the position of the recognized speaker's face. The near-end speaker identification unit 102 can also find the distance from the speaker based on the size of the bounding box. The flash memory 14 stores a table or a function indicating the relationship between the size of the bounding box and the distance in advance. The near-end speaker identification unit 102 can also find the distance from the speaker by comparing the size of the bounding box that has been set with the table stored in the flash memory 14.

[0024] The near-end speaker identification unit 102 obtains the two-dimensional coordinates (X, Y coordinates) of the set bounding box and the distance from the speaker as position information of the speaker. The conversation history recording unit 103 also records the position information of the speaker.

[0025] Fig. 3 shows a configuration in which a speaker is identified based on a far-end video signal and an audio signal and recorded as a conversation history, and Fig. 6 shows the operation in this case. The voice acquisition unit 100 receives a far-end audio signal via the communication unit 19 (S21). The image acquisition unit 101 receives a far-end speaker image via the communication unit 19 (S22). The far-end speaker identification unit 104 identifies the far-end speaker based on the far-end audio signal and the far-end speaker image (S23). The conversation history recording unit 103 records the identifiers indicating the speakers identified by the far-end speaker identification unit 104 and their order as a conversation history (S24).

[0026] The method of speaker identification by the far-end speaker identification unit 104 is the same as the method of speaker identification by the near-end speaker identification unit 102. The far-end speaker identification unit 104 identifies the speaker based on, for example, a speech feature. Alternatively, the far-end speaker identification unit 104 may identify the speaker by performing face recognition processing on the speaker from the received far-end speaker image.

[0027] If the far-end device is capable of identifying a speaker in the same manner as the local device, the far-end speaker identifying unit 104 may receive information identifying the speaker from the far-end device. In this case, the far-end speaker identifying unit 104 can identify the far-end speaker based on the received information.

[0028] Further, the near-end speaker identifying unit 102 and the far-end speaker identifying unit 104 do not necessarily need to identify multiple utterances by one speaker as a single speaker (utterances by the same speaker). For example, the near-end speaker identifying unit 102 and the far-end speaker identifying unit 104 may identify utterances with similar voice features as different speakers. The near-end speaker identifying unit 102 and the far-end speaker identifying unit 104 may identify the same speaker as different speakers if the voice features are slightly different. Even if the near-end speaker identifying unit 102 and the far-end speaker identifying unit 104 identify the same speaker as different speakers and record the conversation history, the prediction unit 51 can predict the next speaker. In other words, the near-end speaker identifying unit 102 and the far-end speaker identifying unit 104 may identify a speaker based on elements that are excessively subdivided to the extent that the next speaker can be predicted or to the extent that one speaker is identified as a different speaker. In addition, the near-end speaker identification unit 102 and the far-end speaker identification unit 104 may record the speaker's voice features themselves, the direction of arrival of the speaker's voice itself, the speaker's facial image itself, or the speaker's position information itself in the conversation history as a result of speaker identification.

[0029] In this embodiment, an example is shown in which there is one far-end device and one-to-one communication is performed, but the sound signal processing device 1 communicates with multiple far-end devices and functions as a multi-point communication system. When communicating with multiple far-end devices, the far-end speaker identifying unit 104 identifies all far-end speakers by performing a process of identifying a speaker for each far-end device.

[0030] FIG. 9 is a diagram showing an example of a conversation history. The conversation history recording unit 103 records the conversation history in chronological order. For example, in the example of FIG. 9, the near-end speaker A1 speaks from time t1 to time t2, and the far-end speaker B1 speaks from time t2 to time t3. The near-end speaker A2 speaks from time t3 to time t4, and the far-end speaker B2 speaks from time t4 to time t5. The near-end speaker A1 speaks from time t5 to time t6, and the far-end speaker B1 speaks from time t6 to time t7. The near-end speaker A2 speaks from time t7 to time t8, and the far-end speaker B1 speaks from time t8 to time t9. After that, the near-end speaker A4 speaks from time t9. It is not essential to record the time information. The conversation history recording unit 103 may simply record an identifier indicating who the speaker was and the order (number) of the speaker.

[0031] The sound signal processing device 1 predicts the next speaker to speak based on the conversation history. The sound signal processing device 1 functions as a speaker prediction device. Fig. 4 shows a functional configuration of the sound signal processing device 1 when making a prediction, and Fig. 7 shows the operation in that case. The example in Fig. 7 shows the prediction operation (operation of the speaker prediction method) when the speaker on the far-end side is currently speaking.

[0032] The voice acquiring unit 100 receives a far-end sound signal via the communication unit 19 (S31). The image acquiring unit 101 also receives a far-end talker image via the communication unit 19 (S32). The far-end talker identifying unit 104 identifies the far-end talker based on the far-end sound signal and the far-end talker image (S33). If the far-end device is a device that can identify a talker in the same way as the device itself, the far-end talker identifying unit 104 may receive information identifying the talker from the far-end device. In this case, the far-end talker identifying unit 104 can identify the far-end talker based on the received information.

[0033] Thereafter, the prediction unit 51 refers to the conversation history recorded in the conversation history recording unit 103 (S34) and predicts the speaker who will speak next (S35). Specifically, the prediction unit 51 detects a part where the far-end speaker identified by the far-end speaker identification unit 104 spoke from the conversation history, and predicts the speaker who will speak next according to the speech probability of the speaker who will speak next to the detected part. For example, when the far-end speaker identification unit 104 identifies the far-end speaker B1, the prediction unit 51 detects times t2 to t3, t6 to t7, and t8 to t9 in the conversation history shown in FIG. 9. Then, the prediction unit 51 obtains the speech probability of each speaker based on the history of the speaker who will speak next at these times. In the example of FIG. 9, after the speaker B1 speaks, the speaker A2 speaks twice, and the speaker A4 speaks once. Therefore, the probability of speaker A2 speaking is 66.7%, and the probability of speaker A4 speaking is 33.3%. The prediction unit 51 predicts that the speaker who will speak next is A2, according to the speech probabilities.

[0034] Furthermore, when the near-end speaker identification unit 102 or the far-end speaker identification unit 104 has recorded the speaker's voice feature value itself, the speaker's voice feature value itself, the speaker's voice arrival direction itself, the speaker's face image itself, or the speaker's position information itself, as a speaker identification result in the conversation history, the prediction unit 51 searches the conversation history recording unit 103 for a speaker identification result close to the current speaker identification result. Then, the prediction unit 51 may predict the next speaker to speak from the speaker identification results recorded in the conversation history recording unit 103 based on a plurality of speaker identification results recorded next to the recording time of the searched current speaker identification result. In this case, the prediction unit 51 may, for example, use the average of a plurality of speaker identification results recorded next to the recording time of the searched current speaker identification result as the speaker prediction result. Alternatively, the prediction unit 51 may use the speaker identification result that has the smallest sum of the difference amount from other speaker prediction results as a representative value as the speaker prediction result.

[0035] The prediction unit 51 may further predict the next speaker based on an image received by the communication unit 19 or an image captured by the camera 11. For example, the prediction unit 51 predicts that a person who has raised their hand is the next speaker. Alternatively, the prediction unit 51 predicts that a person who has their eyes wide open is the next speaker. The prediction unit 51 may also predict the next speaker based on a sound signal received by the communication unit 19 or a sound signal acquired by the microphones 18A to 18F. For example, when the prediction unit 51 recognizes a specific utterance such as "ah" or "eh," it predicts that the corresponding speaker will speak next. However, the prediction unit 51 prioritizes predictions based on conversation history. When there are multiple speakers with the same or similar speaking probability (for example, within 10%), it is preferable that the prediction unit 51 predicts the next speaker based on the received image.

[0036] Thereafter, the camera image control unit 105 controls the image captured by the camera 11 in accordance with the prediction result of the prediction unit 51 (S36). The image control is, for example, a framing process. The camera image control unit 105 performs a framing process to cut out a face image of the speaker A4 from the image of the camera 11 acquired by the image acquisition unit 101. The position of each speaker is recorded in the conversation history recording unit 103. The camera image control unit 105 cuts out an image at a position corresponding to the predicted speaker. The camera image control unit 105 outputs the image after the framing process to the communication unit 19. The communication unit 19 transmits the image after the framing process to the far-end side.

[0037] As a result, a face image of the speaker A2 is displayed on a display (not shown) on the far-end side. Therefore, the speaker B1 on the far-end side can speak while looking at the speaker who is most likely to speak next.

[0038] In the above, the framing process performed by the camera image control unit 105 to cut out an image of the predicted speaker from the image captured by the camera 11 has been described, but the camera image control unit 105 may also perform a process to blur positions other than the predicted speaker, for example. Also, the camera image control unit 105 may control the shooting direction and zoom of the camera 11 to capture an enlarged image of the face of the predicted speaker.

[0039] When the near-end speaker identification unit 102 identifies a speaker on the near-end side, the camera image control unit 105 performs a framing process to cut out the identified speaker on the near-end side. Therefore, an image of the speaker currently speaking is transmitted to the device on the far-end side.

[0040] However, the prediction unit 51 may predict the next speaker to speak even when the near-end speaker identification unit 102 identifies a near-end speaker. The camera image control unit 105 may cut out an image of the speaker predicted to speak next, and transmit it to the far-end device via the communication unit 19. When the speaker predicted by the prediction unit 51 to speak next is a near-end speaker, the camera image control unit 105 performs a framing process to cut out a facial image of the near-end speaker from the image of the camera 11. The camera image control unit 105 transmits the cut-out image of the near-end speaker to the far-end device.

[0041] When the speaker predicted by prediction unit 51 to be the next speaker is the far-end speaker, camera image control unit 105 performs framing processing to cut out a facial image of the far-end speaker from the image received via communication unit 19. Camera image control unit 105 displays the cut-out image of the far-end speaker on a display (not shown).

[0042] However, when transmitting the extracted image of the near-end speaker to the far-end device, it is preferable that the camera image control unit 105 periodically switches between an image of the near-end speaker currently speaking and an image of the near-end speaker predicted to speak next, and transmits the image to the far-end device. For example, the camera image control unit 105 transmits an image of the near-end speaker predicted to speak next about once every 10 seconds. Alternatively, the camera image control unit 105 may combine a small image of the speaker predicted to speak next into the image of the currently speaking speaker, and transmit the image to the far-end device.

[0043] The sound signal processing device 1 may perform sound signal processing of a sound signal acquired by a microphone based on a prediction result of the prediction unit 51. Fig. 10 is a block diagram showing a functional configuration of the sound signal processing device 1 when performing sound signal processing based on a prediction result. The sound signal processing device 1 functionally includes a filter generation unit 52 and a sound signal processing unit 53. Furthermore, the sound signal processing unit 53 functionally includes a beamforming processing unit 501, a gain correction unit 502, and an equalizer 503. These functional configurations are realized by the CPU 12 and the DSP 13.

[0044] The filter generation unit 52 generates a correction filter based on the prediction result of the prediction unit 51. The correction filter includes at least the filtering process in the beamforming processing unit 501.

[0045] The beamforming processor 501 performs beamforming by filtering and synthesizing the sound signals acquired by the microphones 18A to 18F. The signal processing related to the beamforming may be any method such as a delay sum method, a Griffiths Jim type, a Sidelobe Canceller type, or a Frost type Adaptive Beamformer.

[0046] The filter generating unit 52 generates filter coefficients that form directivity toward the position of a speaker predicted to speak next, and sets the filter coefficients in the beamforming processing unit 501. This enables the sound signal processing device 1 to acquire the voice of the speaker predicted to speak next with a high S / N ratio from the beginning of the speech.

[0047] Furthermore, the filter generation unit 52 sets the gain of the gain correction unit 502 based on the prediction result of the prediction unit 51. The voice of a speaker is attenuated more as the distance from the microphone increases. Therefore, the filter generation unit 52 generates a gain correction filter that increases the level of the sound signal as the distance from the speaker predicted to speak next increases, and sets the gain correction filter in the gain correction unit 502. This allows the sound signal processing device 1 to acquire the voice of the speaker predicted to speak next at a stable level regardless of the distance from the speaker.

[0048] Furthermore, the filter generating unit 52 may set the frequency characteristics of the equalizer 503 such that the higher the level of the high frequencies, the greater the distance from the speaker predicted to speak next. The voice of a speaker tends to attenuate more in the high frequencies than in the low frequencies as the distance from the microphone increases. Therefore, the sound signal processing device 1 can acquire the voice of the speaker predicted to speak next with stable sound quality regardless of the distance from the speaker.

[0049] The description of the present embodiment should be considered as illustrative in all respects and not restrictive. The scope of the present invention is indicated by the claims, not by the above-described embodiments. Furthermore, the scope of the present invention includes the scope equivalent to the claims. [Explanation of symbols]

[0050] 1...Sound signal processing device 11. Camera 12…CPU 13…DSP 14. Flash memory 15…RAM 16...User I / F 17…Speaker 18A~18F…Microphone 19…Communications Department 51…Prediction Department 52...Filter generation section 53...Audio signal processing unit 100…Audio acquisition unit 101...Image acquisition unit 102…Near-end speaker identification unit 103…Conversation History Recording Section 104…Far-end speaker identification unit 105...Camera image control unit 501...Beamforming processing unit 502...Gain correction unit 503...Equalizer

Claims

1. Acquire the voices of multiple speakers, recording a conversation history of the plurality of speakers; 1. A method for speaker prediction, comprising: The conversation history includes conversation histories of far-end and near-end speakers; Get the audio, Identifying at least the voice of the far-end speaker among the speakers of the acquired voices; predicting a near-end speaker who will next speak from among the plurality of speakers based on the identified voice of the far-end speaker and the conversation history; Speaker prediction methods.

2. Controlling the images captured by the camera based on the results of the prediction. The speaker prediction method according to claim 1 .

3. performing sound signal processing on the sound signal acquired by the microphone based on the result of the prediction; The speaker prediction method according to claim 1 or 2.

4. The control of the image includes a framing process. The speaker prediction method according to claim 2 .

5. The sound signal processing includes beamforming processing. The speaker prediction method according to claim 3.

6. identifying a speaker of the acquired voice based on a voice feature of the acquired voice; The speaker prediction method according to any one of claims 1 to 5.

7. Estimate the direction of arrival of the voice, Identifying a speaker of the acquired voice based on the direction of arrival of the voice. The speaker prediction method according to any one of claims 1 to 6.

8. Get the camera image, Identifying a speaker of the captured voice based on an image from the camera; A method for speaker prediction according to any one of claims 1 to 7.

9. Detecting a portion where the identified speaker spoke from the conversation history; predicting the next speaker according to a speech probability of the speaker who will speak next to the detected portion; The speaker prediction method according to any one of claims 1 to 8.

10. A speech acquisition unit that acquires speech from a plurality of speakers; a conversation history recording unit that records conversation histories of the multiple speakers; A speaker prediction device comprising: the conversation history includes conversation histories of far-end and near-end speakers; The voice acquisition unit acquires voice, a speaker identification unit that identifies the voice of at least the far-end speaker among the speakers of the acquired voices; a prediction unit that predicts a near-end speaker who will next speak from among the plurality of speakers based on the identified voice of the far-end speaker and the conversation history; A speaker prediction device comprising:

11. A camera image control unit controls an image captured by the camera based on the result of the prediction. The speaker prediction device according to claim 10.

12. and a sound signal processing unit that performs sound signal processing on a sound signal acquired by a microphone based on a result of the prediction. The speaker prediction device according to claim 10 or 11.

13. The control of the image includes a framing process. The speaker prediction device according to claim 11.

14. The sound signal processing includes beamforming processing. The speaker prediction device according to claim 12.

15. The speaker identification unit identifies a speaker of the acquired voice based on a voice feature of the acquired voice. The speaker prediction device according to any one of claims 10 to 14.

16. The speaker identification unit estimates a direction from which the voice comes, and identifies a speaker of the acquired voice based on the direction from which the voice comes. The speaker prediction device according to any one of claims 10 to 15.

17. An image acquisition unit that acquires an image from a camera, The speaker identification unit identifies a speaker of the acquired voice based on an image of the camera. The speaker prediction device according to any one of claims 10 to 16.

18. the prediction unit detects a portion where the specified speaker has spoken from the conversation history, and predicts the speaker who will speak next according to an utterance probability of the speaker who will speak next to the detected portion. The speaker prediction device according to any one of claims 10 to 17.

19. A communication unit that receives the voice of a speaker from the far-end side; A microphone for capturing the voice of a near-end speaker; a conversation history recording unit that records a conversation history of a plurality of speakers including the voices received by the communication unit and the voices captured by the microphone; A communication system comprising: the conversation history includes conversation histories of far-end and near-end speakers; a speaker identification unit that identifies the voice of the speaker of the voice on the far-end side received at least by the communication unit; a prediction unit that predicts a near-end speaker who will next speak from among the plurality of speakers based on the identified voice of the far-end speaker and the conversation history; A communication system equipped with

Citation Information

Patent Citations

  • Person detecting method and device equipped with person detecting function

    JP2001067098A

  • Video teleconference terminal

    JP2007243854A

  • Audio processing system

    JP2012029209A

  • Video processing apparatus, video processing method and video processing system

    JP2017034658A

  • Communication system, communication device, and communication program

    JP2017118364A