JP2005338286A - Object sound processor and transport equipment system using same, and object sound processing method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an object sound processor capable of processing an object sound adaptively to different situations depending upon a position in transport equipment, and a transport equipment system using the same. <P>SOLUTION: A speaker as a member of the crew of a ship operates the ship by speaking to microphones 41 and 42 of a speech indication device 10. Speech signals from the microphones 41 and 42 are subjected to speech recognition processing after noise removal processing, and ship operation data are generated based upon recognition results. At a plurality of places of the ship, IC tags 61 to 6M where position identification information is recorded are arranged. An IC tag reader 46 reads the position identification information of the IC tags 61 to 6M and specifies the position of the speaker. Attribute information and environment information of the ship, on the other hand, are received by a communication device 44 and a communication processing section 76. According to the position identification information, attribute information, and environment information, noise removing algorithm is selected and parameters of the speech recognition processing are selected. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a target sound processing apparatus suitable for use in a relatively large transport device such as a ship, a transport device system using the same, and a target sound processing method. A typical example of the target sound is a voice uttered by a person, and a typical example of the target sound process is a voice recognition process for recognizing the voice.

The voice recognition device is used for an operation of an in-vehicle navigation system (see Patent Document 1 below), a ship operation (see Patent Document 2 below), and the like.
Such a speech recognition apparatus includes a microphone that receives a voice uttered by a speaker, an analog / digital conversion circuit that converts an analog speech signal output from the microphone into a digital speech signal, and the analog / digital conversion circuit. And an audio processing circuit for processing the generated digital audio signal.

The speech processing circuit here includes functionally a noise removing unit, an utterance section detecting unit, an acoustic analyzing unit, and a matching unit. The noise removing unit removes components corresponding to ambient noise other than the voice from the voice signal. The utterance section detection unit extracts a signal of a section where the speaker uttered from the digital voice signal. The acoustic analysis unit generates a feature vector for speech recognition processing from the extracted signal of the utterance section. The collation unit collates the generated feature vector with a recognition dictionary and an acoustic model prepared in advance, and generates speech recognition data representing a recognition result.
JP 2003-108191 A Japanese Patent No. 2668450 Japanese Utility Model Publication 8-1552 Yoshikazu Tada, “Study on Construction of Sound Reception System Using Multiple Wearable Microphones” Proceedings of the Acoustical Society of Japan, March 2003, pp.177-178 Nobukatsu Takai, “Introduction to MATLAB for“ Signal Processing ”and“ Image Processing ”” Engineering, 2000 M. Omologo, P. Svaizer et al., "Acousticevent localization using a crosspower-spectrum phase based technique", Proc. ICASSP94, 1994, pp.274-276 SFBOLL, "Suppression of acoustic nise in speech using spectral subtraction", IEEE Thrans. ASSP, vol.27, no.2, April 1979, pp.113-120

In a large transport device such as a ship, a passenger moves in the ship. Therefore, it is convenient if an occupant can carry a portable device capable of performing voice recognition processing and can give a ship operation instruction by wireless communication from the portable device (see, for example, Patent Document 3).
However, in a large transport device, the noise environment varies greatly depending on the position of the occupant who gives voice instructions. For example, in a ship, noise caused by engine sound, wind noise, and the like is completely different depending on whether the occupant is in a cabin, a fly deck, or an out deck.

  There is no proposal for a speech recognition process that assumes such a large change in the noise environment. Therefore, a conventional speech recognition device is not necessarily suitable for use in a large-sized transport device. That is, the voice recognition rate varies greatly depending on the location of the user in the transport equipment, and depending on the location of the user, the voice of the speaker (user) cannot always be recognized with high probability.

Not only speech recognition but also a similar problem occurs when processing a sound signal including a predetermined target sound.
Accordingly, an object of the present invention is to provide a target sound processing apparatus capable of processing a target sound in accordance with different situations depending on the position in the transport equipment, and a transport equipment system using the target sound processing apparatus.

  Another object of the present invention is to provide a target sound processing method capable of processing a target sound in accordance with different situations depending on the position in the transport device.

  In order to achieve the above object, an invention according to claim 1 is a target sound processing apparatus for processing a predetermined target sound generated in a transport device, and specifies a generation position of the target sound in the transport device. Sound signal processing that can input sound position processing means and a sound signal that is an electric signal of a sound including the target sound, and that can perform sound signal processing to process the sound signal and change parameters of the sound signal processing And a parameter switching means for switching a parameter of the sound signal processing in the sound signal processing means in accordance with the generation position specified by the generation position specifying means.

According to this configuration, the generation position of the target sound in the transport device is specified, and the sound signal processing parameters are switched according to the specified generation position. Thereby, the processing mode of the target sound can be adapted to a situation (for example, a noise situation) that varies depending on the position in the transport device.
In addition, since complicated processing is not required for parameter switching, high-accuracy sound signal processing can be realized with low load processing.

  The transport device is a transport device that is so large that the sound signal processing accuracy is substantially affected by switching the sound signal processing parameters in accordance with the generation position of the target sound in the transport device. More specifically, in the case where the target sound is a sound emitted from the user (for example, a voice of the user), it is a transport device in which the user (occupant) can move. Examples of such transportation equipment are represented by marine transportation equipment represented by ships, ground transportation equipment represented by trains and large vehicles (for example, buses), aerial transportation equipment represented by airplanes, and spacecrafts. Includes space transportation equipment.

Examples of sounds to be processed include sounds emitted from human bodies such as voices, applause or footsteps, sounds emitted from sound generators such as musical instruments and other sound sources (such as whistles, horns, and electronic sound sources) A mechanical sound emitted from a machine (such as an engine or a steering device) provided in a transportation device can be given.
Examples of sound signal processing include target sound recognition processing (for example, human speech recognition processing), processing for detecting a specific sound, and noise removal processing.

Examples of the target sound input source include a microphone that receives a sound including the target sound and converts it into a sound signal, and a playback device that reproduces a pre-recorded sound.
The generation position specifying means may be any means that specifies a plurality of separated positions having different sound environments so that the sound signal processing parameters need to be switched. For example, when the transport device is a ship, the generation position specifying unit may specify the generation position of the target sound in any one of a cockpit, a cabin, a flying deck, and an out deck.

  Further, the generation position specifying means may distinguish and specify a plurality of positions that are substantially different in influence on sound signal processing by a main noise source. More specifically, the generation position specifying means may identify and specify a plurality of positions having different distances from a main noise source (for example, a prime mover). Further, for example, when the main noise source is wind noise generated according to the movement of the transport device, the generation position specifying means may identify and specify a plurality of positions having different wind-contact situations. Furthermore, the generation position specifying means may identify and specify the inside of the shielded space (for example, in the cabin of the ship) and the outside of the shielded space (open space, for example, the deck of the ship).

  The target sound processing device is preferably configured in a form that the user can carry or wear when processing the sound generated by the user himself or the sound generated in the vicinity of the user. For example, a voice instruction device that recognizes a user's voice and operates a device (a main device for maneuvering the transportation device or an auxiliary device other than the maneuvering device) provided in the transportation device based on the recognition result However, this target sound processing apparatus may be applied.

  The invention according to claim 2 further includes transport equipment information acquisition means for acquiring transport equipment information including at least one of attribute information of the transport equipment and environmental information of the transport equipment, and the parameter switching means includes the generation The sound signal processing parameter in the sound signal processing means is switched based on the transport equipment information acquired by the transport equipment information acquisition means in addition to the generation position specified by the position specifying means. The target sound processing device according to claim 1.

According to this configuration, the sound signal processing parameters are switched in consideration of the attribute information and / or environmental information of the transport equipment. Thereby, the accuracy of the sound signal processing can be further improved.
The attribute information of the transport equipment is information on the internal properties of the transport equipment (particularly related to the noise environment). The attribute information of the transport device can be further divided into static attribute information unique to the transport device and dynamic attribute information that varies during the operation period of the transport device. Examples of static attribute information include the size of the transport device, the type of transport device, and the type of prime mover (transport device built-in type, externally mounted type, etc.). Examples of the dynamic attribute information include the rotational speed of the prime mover and the state of the prime mover. The prime mover for obtaining the propulsive force of the transportation device may be a heat engine (engine) using thermal energy or an electric motor using electric energy. By configuring so as to acquire such attribute information of the transport device, it is possible to cope with different situations for each type of transport device.

The environmental information of the transport equipment is information on external environmental factors surrounding the transport equipment (particularly external factors related to the noise environment). Examples of environmental information of the transport device include wind intensity, wind direction, and wave magnitude around the transport device. Moreover, as an example of the environmental information of the space transportation equipment, information on whether the transportation equipment is in the atmosphere or outside the atmosphere can be given.
The transport device information acquisition means may acquire all of the exemplified attribute information and environment information, or may acquire a part of them.

According to a third aspect of the present invention, the transport device includes a database in which attribute information (particularly static attribute information) of the transport device is stored, and the transport device information acquisition unit is configured to receive the transport device from the database. 3. The target sound processing apparatus according to claim 2, further comprising attribute information acquisition means for acquiring the attribute information.
According to this configuration, since the attribute information of the transport device can be acquired using the database that is the facility of the transport device, accurate attribute information of the transport device can be obtained with a simple configuration.

For example, the transport device may include a LAN (Local Area Network) inside and the database may be connected to the LAN. In this case, the attribute information of the transport equipment can be acquired from the database by providing the target sound processing device with communication means for communication with the LAN.
The attribute information of the transport device may be input to the target sound processing device by an input operation from an input device such as a keyboard.

According to a fourth aspect of the present invention, the transport equipment includes transport equipment information detection means for detecting the transport equipment information, and the transport equipment information acquisition means is transport equipment detected by the transport equipment information detection means. 4. The target sound processing apparatus according to claim 2, wherein the information is acquired from the transport device.
With this configuration, the attribute information and / or environmental information of the transport device can be acquired by using the transport device information detection means that is a facility of the transport device. Therefore, the transport equipment information can be acquired without complicating the configuration of the target sound processing device itself.

The transport equipment information detecting means may include attribute information detecting means for detecting the attribute information (particularly dynamic attribute information). Examples of such attribute information detection means include sensors that detect the operating state of the equipment provided in the transport equipment, such as a motor speed sensor.
In addition, the transport equipment information detection means includes environmental information detection means for detecting environmental information around the transport equipment, and the transport equipment information acquisition means uses the environmental information detected by the environmental information detection means. It may include environmental information acquisition means that acquires from the transport equipment. Examples of the environmental information detection means include an anemometer, an anemometer, and a wave height meter. The environmental information detection means may acquire environmental information around the transport device by information communication.

For example, when the communication device such as the LAN is provided in the transportation device, the target sound processing device includes a device (communication device) for acquiring the transportation device information via the LAN. Preferably it is.
According to a fifth aspect of the present invention, the sound signal processing means includes noise removal processing means for performing noise removal processing for removing a noise component from the input sound signal, and the parameter switching means is the noise removal processing means. 5. The object according to claim 1, further comprising noise removal parameter switching means for switching a noise removal parameter (algorithm and / or coefficient of the noise removal process) applied to the noise removal process executed by. It is a sound processing device.

According to this configuration, the parameter of the noise removal process is switched according to the target sound generation position in the transport device. Thereby, it is possible to perform a noise removal process suitable for noise situations at a plurality of different positions in the transport device, and to extract the component of the target sound from the sound signal with high accuracy.
Further, in the case of the combination with the feature of claim 2, since the noise removal parameter is selected by further adding the attribute information and / or environment information of the transport equipment, more appropriate noise removal processing is performed. Is called.

  Examples of the noise removal parameters include types of noise removal algorithms, combinations of a plurality of types of noise removal algorithms, and coefficients used for the noise removal algorithms. That is, the selection of the noise removal parameter may be selection of one noise removal algorithm from a plurality of types of noise removal algorithms. Further, the selection of the noise removal parameter may be a selection of a combination of two or more noise removal algorithms from a plurality of types of algorithms. Further, the selection of denoising parameters may be a selection of coefficients that are applied to a particular denoising algorithm.

Examples of the noise removal algorithm include high frequency enhancement filter processing and spectral subtraction processing.
According to a sixth aspect of the present invention, the sound signal processing means includes target sound extraction means for executing target sound section extraction processing for extracting a signal section including the target sound from an input sound signal, and the parameter switching means is 6. The target sound according to claim 1, further comprising target sound extraction parameter switching means for switching a target sound extraction parameter applied to the target sound section extraction processing executed by the target sound extraction means. It is a processing device.

  According to this configuration, the parameter of the target sound segment extraction process is switched according to the target sound generation position in the transport device. Thereby, the signal section containing the target sound can be accurately extracted. In particular, in the case of the combination with the configuration of claim 2, the target sound extraction parameter is changed by further adding the attribute information and / or environment information of the transport equipment, so that the signal section of the target sound is made more accurate. Can be extracted.

When the target sound is human speech, the target sound section extraction process may be an utterance section extraction (detection) process that extracts a signal of the utterance section uttered by the speaker.
For example, the target sound segment extraction process is a process of extracting the signal segment of the target sound from the sound signal by analyzing the frequency spectrum of the input sound signal and performing statistical processing on the analysis result. Good. In this case, the parameter applied to the statistical processing may be switched according to the target sound generation position in the transport device and the attribute information / environment information of the transport device.

The sound signal processing means has a plurality of types of acoustic models that can be collated with the sound signal of the target sound, and the parameter switching means is an acoustic model that switches between the plurality of types of acoustic models. 7. The target sound processing apparatus according to claim 1, further comprising switching means.
According to this configuration, the acoustic model is switched according to the target sound generation position in the transport device. Thereby, the collation process with an object sound and an acoustic model can be performed correctly, for example, the precision of the recognition process of an object sound can be improved.

In particular, in the case of the combination with the configuration of claim 2, the acoustic model is changed by further adding the attribute information and / or environment information of the transport equipment, so that more accurate matching processing is possible.
The invention according to claim 8 is characterized in that the transport device includes a plurality of identification information carriers arranged (recorded) at different positions in the transport device and each carrying (recording) position identification information. 8. The target sound processing apparatus according to claim 1, wherein the specifying unit includes a reading device that reads position identification information carried on the identification information carrier.

According to this configuration, the position where the target sound is generated is specified by reading the position identification information carried by the identification information carrier disposed at a plurality of positions of the transport device by the reading device.
The identification information carrier may be disposed in the vicinity of the target sound generation position. For example, when the target sound generation position (speech position) moves around in the transport device, such as when the target sound is the voice of the user, the user carries identification information in the vicinity of the place where the target speech is to be made. The reader is made to read the body position identification information. Thereby, the parameter corresponding to the user's utterance position is set.

Examples of the identification information carrier include information recording media such as IC tags and barcode labels. When using an IC tag, an IC tag reader is used as the reader. Similarly, when a barcode label is used, a barcode reader is used as the reading device.
Moreover, it is preferable that the reading device or the entire target sound processing device including the reading device is configured to be carried or worn by the user.

The generation position specifying means may include position information input means for a user to input position information. According to this configuration, since the user inputs the position information, a highly accurate sound signal process adapted to the target sound generation position can be realized with a simple configuration. An input device such as a keyboard can be applied as the position information input means.
The invention according to claim 9 is characterized in that the transport device is a ship, and the generation position specifying means specifies the generation position of the target sound in the ship. The target sound processing device according to claim 1.

In general, in ships, the generation positions of target sounds such as cabins and decks are widely distributed, and the noise environment varies greatly depending on the distance from the prime mover and the wind conditions. Therefore, the sound signal processing with high accuracy can be performed by specifying the generation position of the target sound in the ship and switching the sound signal processing parameters according to the specified generation position.
In the invention according to claim 10, the target sound is a voice uttered by a speaker in the transport device,
10. The target sound processing apparatus according to claim 1, wherein the sound signal processing means includes voice recognition processing means for executing voice recognition processing for recognizing the voice of the speaker.

According to this configuration, the utterance position of the speaker in the transport device is specified, and the parameters applied to the speech recognition process are switched according to the specified utterance position. Thereby, the voice recognition rate can be improved.
The speaker's voice may be input from a handset or headset microphone. Further, the microphone and the target sound processing device may be integrated in a portable housing. When the transport device is a ship, the microphone for inputting the voice of the speaker may be attached to a wearing tool such as a life jacket worn by the occupant. The target sound processing device may be attached to the mounting tool. In order to obtain a high speech recognition rate, it is preferable that the positional relationship between the utterance position and the microphone is kept constant. Therefore, a headset type microphone or a microphone attached to a wearing device is used. It is preferable.

  Further, a plurality of microphones for inputting a speaker's voice may be distributed at a plurality of positions in the transport device. In this case, for example, the speaker holds an authentication information carrier (for example, a card-like one, an IC tag, a bar code label, etc.) carrying predetermined authentication information, and a reading device is placed near each microphone installation location. Deploy. In use, the speaker causes the reader to read the authentication information of the authentication information carrier. As a result, the position of the speaker can be specified depending on where the authentication information is input by the reading device. Based on the specified position information, the parameters applied to the speech recognition process may be switched.

The invention described in claim 11 is a transport apparatus system comprising the transport apparatus and the target sound processing device according to any one of claims 1 to 10. With this configuration, it is possible to improve the accuracy of processing the target sound on the transport device.
According to a twelfth aspect of the present invention, an instruction for operating the transport device by the speaker is based on a voice recognition result by the transport device, the target sound processing device according to the tenth embodiment, and the voice recognition processing means. A transport equipment operation data generating means for identifying and generating transport equipment operation data corresponding to the instruction, and a device provided in the transport equipment based on the transport equipment operation data generated by the transport equipment operation data generating means A transportation device system including a device control means to be operated.

  With this configuration, transportation device operation data corresponding to the voice instruction given by the speaker is generated, and based on this, various devices provided in the transportation device can be operated. Since the parameters applied to the speech recognition process are appropriately switched according to the position of the speaker, a high speech recognition rate can be ensured. Therefore, the transportation device can be operated smoothly by voice instruction.

The transport equipment operation data generation means may be provided on the target sound processing apparatus side provided with voice recognition means. Further, the transport equipment operation data generating means may be provided on the transport equipment side.
A thirteenth aspect of the present invention is a target sound processing method for processing a predetermined target sound generated in a transport device, the method including a step of specifying a generation position of the target sound in the transport device and the target sound Performing sound signal processing on a sound signal that is an electric signal of sound, and switching a parameter applied to the sound signal processing in accordance with the specified generation position of the target sound. This is a characteristic target sound processing method.

By this method, the processing mode of the target sound can be applied to a situation that varies depending on the position in the transport device. Therefore, the target sound can be processed with high accuracy while in the transport device.
The invention according to claim 14 further includes a step of acquiring transport device information including at least one of attribute information of the transport device and environmental information of the transport device, and the step of switching the parameter includes the specified object The target sound processing method according to claim 13, further comprising a step of switching a parameter applied to the sound signal processing based on the acquired transport device information in addition to a sound generation position.

According to this method, the sound signal processing parameters are switched in consideration of the attribute information and environmental information of the transport equipment, so that the accuracy of the sound signal processing can be further improved.
With respect to the inventions of these methods, the modifications described in relation to the above-described target sound processing apparatus can be made.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.
FIG. 1 is an illustrative view showing a conceptual configuration of a ship system as an example of a transportation device system according to an embodiment of the present invention, and FIG. 2 is an illustrative plan view of the ships constituting the ship system. is there. This ship system can perform operation of equipment provided in the ship 1 which is an example of transport equipment by voice instruction. In this embodiment, the operable equipment includes an outboard motor 2 as an example of a propulsion device, a steering device 3 for steering, a signal red flame device 4, a marine telephone device 5, and a rubber boat operating device 7. The operation of the devices provided in these ships 1 is hereinafter referred to as “ship operation”. Therefore, in ship operation, “ship operation” which is an operation of equipment essential for navigation of the ship 1 such as a propulsion device (outboard motor 2) and a steering device (steering device 3), a signal red flame device 4, “Vehicle telephone device 5” and “rubber boat actuating device 7” include “accessory operation” which is not directly related to the navigation of ship 1.

  The voice instruction device 10 for giving a voice instruction for ship operation is of a transceiver type (handset type) carried by a speaker 8 who is an occupant of the ship 1. The voice instruction device 10 performs radio communication with the ship 1 using radio waves. Thereby, the speaker 8 holding the voice instruction device 10 can perform a ship operation by voice instruction. This voice instruction device 10 corresponds to a target sound processing device.

  The marine vessel 1 includes a hull (hull) 51 and an upper structure (superstructure) 52 provided on the hull 51. In the example of FIG. 1, the upper structure 52 is provided in an out deck (outer deck) 53 that covers the upper surface of the hull 51, a cabin 54 provided substantially in the center of the out deck 53, and in front of the cabin 54. And the cockpit 55. The ceiling portion of the cabin 54 is located above the out deck 53 and forms a flying deck 56.

  An outboard motor (outboard engine) 2 is attached to the hull 51 at the tail. In the hull 51, a steering unit 13, a shift / throttle operation unit 14, a key input unit 15, and a wireless communication unit 16 are provided. Further, in the hull 51, the signal red flame device 4, the marine telephone device 5 and the rubber boat operating device 7 are arranged. The steering device 3 includes a steering unit 13, a steering handle 18 coupled to the steering unit 13, a steering cable 19 coupled to the outboard motor 2, and the like.

  As shown in FIG. 2, IC tags 61, 62,... As identification information carriers carrying identification information (position identification information) for specifying each position are provided at a plurality of different positions on the ship 1. .., 6M (where M is a natural number of 2 or more) are arranged. That is, the IC tags 61 to 6 </ b> M are distributed and arranged in a plurality of places where the speaker 8 may be located in the ship 1. More specifically, the IC tag 61 is disposed in the cockpit 55, the IC tag 62 is disposed in the cabin 54, and the IC tag 63 is disposed on the right side (starboard) of the cockpit 55 on the bow side of the out deck 53. The IC tag 64 is disposed on the left side (port) of the cockpit 55 on the bow side of the out deck 53, and the IC tag 6M is disposed on the flying deck 56.

  FIG. 3 is a block diagram mainly showing an electrical configuration inside the ship 1. The ship 1 is provided with an inboard LAN (local area network) 11. The inboard LAN 11 includes a database device 12, a steering unit 13, a shift / throttle operation unit 14, a key input unit 15, a wireless communication unit 16, a processing / output unit 17, a signal red flame device 4, a ship telephone device 5, and a rubber boat operating device. 7 is connected. Further, an environmental information sensor 9 (environment information detecting means) for obtaining environmental information of the ship 1 is connected to the inboard LAN 11. Examples of the environmental information sensor 9 include an anemometer, an anemometer, and a wave height meter (for example, a G sensor). Known hardware standards and protocols can be applied to the inboard LAN 11.

The outboard motor 2 includes an engine 21, an engine control unit 22, and a starter motor 23. The engine control unit 22 is connected to the inboard LAN 11. The engine control unit 22 is provided with an actuator 24 for changing the shift state and the throttle opening state of the engine 21.
The starter motor 23 is supplied with power from the battery 26 via a starter relay 27. The starter relay 27 is actuated by the key input unit 15. By closing the starter relay 27, electric power is supplied to the starter motor 23, and the engine 21 is started. The key input unit 15 closes the starter relay 27 in response to a passenger's manual operation, and closes the starter relay 27 in response to a predetermined control signal given from the inboard LAN 11.

  The database device 12 stores the attribute information of the ship 1 in advance, and outputs the attribute information as necessary. Examples of the attribute information of the ship 1 include the size of the ship 1, the type of propulsion device, the engine speed, and the state of the engine. Among these, static attribute information, such as attribute information indicating the size of the ship 1 and the type of propulsion device, is stored in the database device 12, and the engine speed and engine state (throttle opening, shift) The dynamic attribute information that fluctuates with time such as the engine cover open / closed state) is detected by sensors 20 such as an engine speed sensor.

  Examples of attribute information indicating the type of propulsion device include outboard motors, inboard / outboard motors (inboard engine / outboard drive), inboard motors (inboard drive), and water jet drive. Information on the type can include information on the type of engine such as a 2-stroke gasoline engine, a 4-stroke gasoline engine, and a diesel engine.

  The steering unit 13 includes a handle angle detection unit 31 that detects the handle angle of the steering handle 18 and a steering actuator 32 that generates a mechanical displacement corresponding to the detected handle angle. The mechanical displacement generated by the steering actuator 32 is transmitted to the outboard motor 2 via the steering cable 19. Thereby, the direction of the outboard motor 2 is changed.

  The shift / throttle operation unit 14 transmits shift operation information to the engine control unit 22 via the inboard LAN 11, and throttle information transmission that transmits throttle operation information to the engine control unit 22 via the inboard LAN 11. Part 34. The shift operation and the throttle operation in the shift / throttle operation unit 14 are, for example, operations for changing the tilt angle of the operation lever (remote control lever) 14a (see FIG. 1). When the operation lever 14a is tilted forward by a certain amount, the shift is changed from neutral to forward rotation (forward), and further, the throttle operation can be performed to increase the throttle opening as the operation lever 14a is tilted forward. Further, when the operation lever 14a is tilted backward by a certain amount, the shift is reversed from the neutral position (reverse), and further, the throttle operation can be performed to increase the throttle opening as the operation lever 14a is tilted backward.

  The engine control unit 22 outputs a signal for operating the actuator 24 in accordance with the shift operation and throttle operation information from the shift / throttle operation unit 14. In response to this signal, the actuator 24 generates a mechanical displacement for adjusting the opening of a throttle valve (not shown) of the engine 21 or operating a shift mechanism (clutch, not shown). To do. In the engine 21, its rotational speed is detected by an engine rotational speed sensor, and other information indicating the state of the engine 21 such as the opening degree of the throttle valve and a shift state (clutch state) is detected by various sensors 20. The Detection signals of these sensors 20 including the engine speed sensor are supplied to the engine control unit 22. Thereby, the engine control unit 22 can detect the control state of the engine 21.

  The wireless communication unit 16 communicates with the voice instruction device 10 via the antenna 35 (see FIG. 1). When receiving the vessel operation data from the voice instruction device 10, the wireless communication unit 16 passes the received vessel operation data to the processing / output unit 17 via the inboard LAN 11. Based on the received ship operation data, the processing / output unit 17 transmits a control signal for ship operation to the corresponding device via the inboard LAN 11. That is, the processing / output unit 17 functions as a device control unit that operates a device provided in the ship 1 according to the ship operation data.

  In addition, the wireless communication unit 16 receives a request for transmission of the attribute information and environment information of the ship 1 from the voice instruction device 10, and passes the request to the processing / output unit 17 via the inboard LAN 11. In response to this transmission request, the processing / output unit 17 reads out the static attribute information of the ship 1 from the database device 12 and further acquires dynamic attribute information such as the engine speed from the engine control unit 22. . Further, the processing / output unit 17 acquires environmental information of the ship 1 from the environmental information sensor 9. The acquired attribute information and environment information are transferred from the processing / output unit 17 to the wireless communication unit 16 via the inboard LAN 11, and transmitted from the wireless communication unit 16 to the voice instruction device 10.

  FIG. 4 is a block diagram for explaining the electrical configuration of the voice instruction device 10. The voice instruction device 10 includes first and second microphones 41 and 42 for receiving voices spoken by the speaker 8 in two channels, a communication device 44, an arithmetic processing unit 45, and IC tags 61 to 6M. IC tag reader 46 for reading the position identification information from the inside of the transceiver type housing. The communication device 44 is a wireless communication unit that transmits and receives radio waves to and from the wireless communication unit 16 on the ship 1 side via the antenna 44A.

  The arithmetic processing unit 45 is an electronic system including a microcomputer and a memory. The arithmetic processing unit 45 is connected to the first and second microphones 41 and 42, the communication device 44 and the IC tag reading device 46. The arithmetic processing unit 45 generates voice recognition processing for recognizing the voice of the speaker 8 received by the first and second microphones 41 and 42 and ship operation data to be transmitted from the communication device 44 to the ship 1. The main function is to perform ship operation data generation processing. The arithmetic processing unit 45 realizes these functions by causing a microcomputer to execute a program.

  The arithmetic processing unit 45 includes A / D (analog / digital) converters 71 and 72 that convert analog audio signals output from the first and second microphones 41 and 42 into digital audio signals, and the A / D converter 71. , 72 performs a noise removal processing unit 73 for performing noise removal processing on the speech signal, a speech recognition processing unit 74 for performing speech recognition processing on the speech signal after the noise removal processing, and the speech recognition processing unit 74. A ship operation data generation unit 75 (transport equipment operation data generation means) that generates ship operation data based on the processing result (recognition result) by the communication processing unit 76 that executes data processing for communication via the communication device 44. And.

  The noise removal processing unit 73 removes a noise part included in the voice signal and outputs a voice signal with the noise part reduced. For example, a noise removal method such as a spectral subtraction method or a high-frequency emphasis filter can be applied to the processing in the noise removal processing unit 73. The arithmetic processing unit 45 includes a noise removal algorithm buffer 77 storing a plurality of types of noise removal processing algorithms 731 to 73k (k is a natural number of 2 or more) applicable to the noise removal processing in the noise removal processing unit 73, and the noise removal. A noise removal algorithm switching unit 78 (noise removal parameter switching means) for switching an algorithm applied to the processing is provided. The noise removal algorithm switching unit 78 includes position identification information read by the IC tag reading device 46, information acquired from the inboard LAN 11 by the communication device 44 and the communication processing unit 76 (attribute information and environment information, transport equipment information), and the like. Based on the above, the algorithm applied to the noise removal processing in the noise removal processing unit 73 is switched.

The communication device 44 and the communication processing unit 76 have a function as attribute information acquisition means for acquiring attribute information from the ship 1 and a function as environment information acquisition means for acquiring environment information from the ship 1.
FIG. 5 is a block diagram for explaining a functional configuration of the voice recognition processing unit 74. The voice recognition processing unit 74 extracts a voice signal of the utterance section of the speaker 8 from the two-channel voice signals subjected to the noise removal process, and this utterance section detection. An acoustic analysis unit 92 that analyzes the voice signal extracted by the unit 91 to generate a feature vector representing an acoustic feature, and an acoustic model and a recognition dictionary in which the feature vector generated by the acoustic analysis unit 92 is prepared in advance 98 and a verification unit 93 that generates verification result data.

  The utterance section detection unit 91 can change a parameter (speech detection parameter, target sound extraction parameter) of the utterance section detection process. To switch the utterance detection parameter to a plurality of types based on position identification information read by the IC tag reader 46 and information (attribute information and environment information) acquired from the inboard LAN 11 via the communication device 44 In addition, an utterance detection parameter switching unit 94 is provided. The utterance detection parameter switching unit 94 selects one of a plurality of types of utterance detection parameters 911 to 91m (m is a natural number of 2 or more) stored in the utterance detection parameter buffer 95, and an utterance section detection unit. 91 and has a function as a target sound extraction parameter switching means.

  On the other hand, an acoustic model switching unit 96 is provided to switch the acoustic model to be collated by the collation unit 93. The acoustic model switching unit 96 selects an acoustic model based on the position identification information read by the IC tag reading device 46 and information (attribute information and environmental information) acquired from the inboard LAN 11 via the communication device 44. Switch. The acoustic model switching unit 96 selects one of a plurality of types of acoustic models 931 to 93n (n is a natural number of 2 or more) stored in the acoustic model buffer 97, and performs collation processing in the collation unit 93. Apply.

The collation unit 93 collates the feature vector generated by the acoustic analysis unit 92 with the acoustic model selected by the acoustic model switching unit 96 and further collates with the recognition dictionary 98 to generate recognition result data.
The speaker 8 moves to various positions in the ship 1 and uses the voice instruction device 10 to operate the ship. At this time, the noise environment around the speaker 8 varies depending on the position of the speaker 8 in the ship 1. For example, the engine sound is louder as it approaches the stern, and the out deck 53 and the flying deck 56 are louder than the cockpit 55 or the cabin 54. Further, regarding wind noise, the out deck 53 is larger than the cockpit 55 and the cabin 54, and the flying deck 56 is larger than the out deck 53.

Therefore, in this embodiment, the noise removal processing parameters (algorithms) are switched according to the position of the speaker 8 in the ship 1, and further the speech recognition processing parameters (speech detection parameters and acoustic models) are switched. The speech recognition rate is improved.
Furthermore, the voice recognition rate is also affected by the attributes of the ship 1 and the environmental conditions. The attribute of the ship 1 is the property of the ship 1 represented by the attribute information described above. The environmental situation is the surrounding environment of the ship 1 represented by the above-described environmental information.

  Static attribute information stored in the database device 12, dynamic attribute information detected by sensors 20 such as an engine speed sensor, and environment information detected by the environment information sensor 9 are wirelessly transmitted from the inboard LAN 11. It is given to the communication unit 16. These pieces of information are further received by the communication device 44 and transferred to the noise removal algorithm switching unit 78 and the speech recognition processing unit 74. Based on these attribute information and environment information, parameters (algorithms) for noise removal processing are switched, and parameters (speech detection parameters and acoustic models) for speech recognition processing are switched. Thereby, the speech recognition rate is further improved. Moreover, since the attribute information such as the size of the ship 1 is also acquired, the voice instruction device 10 can easily adapt to a plurality of types of ships 1 and realize a good voice recognition rate.

  When the speaker 8 gives a ship operation instruction by voice, the IC tag reader 46 is held over the IC tags 61 to 6M near the speaker 8 so that the position identification information is read, and then the microphones 41 and 42 are read. Speak the sound for ship operation. As long as the speaker 8 does not move in the ship 1, the speaker 8 can speak to the microphones 41 and 42 without requiring the reading operation of the IC tags 61 to 6M.

  When the position identification information of any of the IC tags 61 to 6M is read, in response to this, the communication processing unit 76 transmits the attribute information and the environment information to the ship 1 via the communication device 44. Give the request to send. This transmission request is given to the wireless communication unit 16 provided in the ship 1. The wireless communication unit 16 acquires attribute information from the database device 12 and the engine control unit 22 via the inboard LAN 11, and further acquires environment information from the environment information sensor 9. Then, the wireless communication unit 16 transmits the attribute information and the environment information to the voice instruction device 10. In the voice instruction device 10, the attribute information and the environment information transmitted from the ship 1 are received by the communication device 44 and transferred to the communication processing unit 76.

  The position identification information, attribute information, and environment information acquired in this way are provided to the noise removal algorithm switching unit 78, the utterance detection parameter switching unit 94, and the acoustic model switching unit 96. Thus, noise removal processing and speech recognition processing that are well adapted to the noise environment and the like are performed, so that a high speech recognition rate can be realized. In addition, since complicated processing for estimating the noise environment from the voice signal is not required, it is possible to immediately adapt the mode of the voice recognition processing to the noise environment with a small processing load.

  FIG. 6 is a block diagram illustrating a configuration example of the noise removal processing unit 73. In this configuration example, the noise removal processing unit 73 has a plurality of noise removal filter processing units. More specifically, the noise removal processing unit 73 includes high-frequency enhancement filter units 81 and 82 as first filter processing units, and spectral subtraction filter units 83 and 84 as second filter processing units. For processing in the spectral subtraction filter units 83 and 84, between the high frequency enhancement filter unit 81 and the spectral subtraction filter unit 83, and between the high frequency enhancement filter unit 82 and the spectral subtraction filter unit 84, respectively. Pre-processing units 89 and 90 are interposed.

  The high-frequency emphasis filter unit 81 and the spectral subtraction filter unit 83 perform filter processing on the audio signal output from the first microphone 41. Similarly, the high-frequency emphasis filter unit 82 and the spectral subtraction filter unit 84 perform filter processing on the audio signal output from the second microphone 42. That is, the two-channel audio signals output from the first and second microphones 41 and 42 are subjected to two types of noise removal filter processing (high-frequency emphasis filter processing and spectral subtraction filter processing). More specifically, the audio signal of each channel is subjected to spectral subtraction filter processing after being subjected to high-frequency emphasis filter processing.

  For the high frequency emphasis filter sections 81 and 82, the coefficient γ for adjusting the validity / invalidity of the filter processing and the characteristics of the filter processing can be variably set. Further, the coefficients δ and ε for adjusting the validity / invalidity of the filter processing and the characteristics of the filter processing can be variably set for the spectral subtraction filter units 83 and 84 as well. That is, in the configuration example of FIG. 6, a plurality of sets of coefficients with one set of coefficients γ, δ, and ε are accumulated in the noise removal algorithm buffer 77, and an appropriate coefficient set is selected from among these sets. The noise removal algorithm switching unit 78 sets the noise removal processing unit 73.

The high frequency emphasizing filter units 81 and 82 execute digital filter processing represented by the following equations (1) and (2), for example.
y ₁ (i) = x ₁ (i) −γx ₁ (i−1) (1)
y ₂ (i) = x ₂ (i) −γx ₂ (i−1) (2)
However, in the equations (1) and (2), the two-channel input signals sampled and digitized by the i-th (i is a natural number) by the A / D converters 71 and 72 are respectively x _1. (i) and x ₂ (i) are represented, and the output signals after filtering are represented as y ₁ (i) and y ₂ (i).

The value of the coefficient γ is often set to 0.90 to 1.00. If this coefficient γ is set to “0”, the high frequency emphasis filter processing is invalidated, and y ₁ (i) = x ₁ (i), y ₂ (i) = x ₁ (i). Further, the filter characteristic can be adjusted by variably setting the value of the coefficient γ.
The spectral subtraction filter units 83 and 84 execute digital filter processing represented by the following expressions (3) and (4), for example.

However, in the equations (3) and (4), X ₁ (ω) and X ₂ (ω) are two-channel observation signals (voice signals after high-frequency emphasis filter processing in the example of FIG. 6). Each represents a power spectrum (frequency function), and S ₁ (ω) and S ₂ (ω) represent output signals after filter processing on them. N (ω) represents the estimated power spectrum of the noise signal. ω represents an angular frequency.

  The coefficient δ is called a subtraction coefficient, and δ ≧ 0. Ε is called a flooring coefficient, and 0 ≦ ε ≦ 1. Depending on the environment around the first and second microphones 41 and 42, the optimum pair of δ and ε is different. If δ = 0 (and ε = 1 if necessary), the spectral subtraction filter processing is invalidated. Also, the filter characteristics can be adjusted by variably setting the values of the coefficients δ and ε.

The spectral subtraction method is disclosed in Non-Patent Document 4, for example.
Table 1 below shows experimental results showing examples of effects of the high-frequency emphasis filter. In Table 1, the spectral subtraction filter units 83 and 84 are disabled, and the high frequency emphasis filter units 81 and 82 are switched between valid (γ = 1) / invalid (γ = 0), and the speech recognition rate (unit%) is set. The measurement results are shown. The measurement was performed for a plurality of values of the engine speed at a plurality of locations in the ship 1.

  Inside the cabin, there is no wind effect and the engine noise is relatively low. The rear of the out deck is located in front of the engine and has the highest engine noise. However, behind the out deck, there is little wind noise because the deck house (the deck room) that forms the cabin and the like serves as a windbreak. The flying deck is far away from the engine compared to the rear of the out deck, but has the strongest wind.

  As understood from Table 1, in a place where the influence of wind is strong (flying deck), the speech recognition rate is better when the high frequency emphasis filter is used. Also, where the influence of wind and engine noise are small (cabin: engine speed = 500,2200, outdeck: engine speed = 500), the presence or absence of the high-frequency emphasis filter is not so relevant. On the other hand, in places where there is almost no wind influence and only engine noise is large (cabin: engine speed = 2800, out deck: engine speed = 2200, 2800), using the high frequency enhancement filter processing, the speech recognition rate Becomes worse.

  Table 2 below shows experimental results showing examples of effects of the spectral subtraction filter. In Table 2, the high-frequency emphasis filter units 81 and 82 are disabled, and the spectral subtraction filter units 83 and 84 are enabled (for example, δ = 2.0, ε = 0.4) / disabled (for example, δ = 0). , Ε = 1), and the result of measuring the voice recognition rate (unit%) is shown. The measurement was performed for a plurality of values of the engine speed at a plurality of locations in the ship 1.

  Noise from the engine can be regarded as almost stationary noise. As can be understood from Table 2, the spectral subtraction filter has an effect in the cabin and the out deck when the engine noise is large (engine speed = 2800). However, in these places, when the engine noise is low (engine speed = 500), the speech recognition rate is lowered. In the flying deck, the effect of the spectral subtraction filter is not seen regardless of the engine speed.

From the experimental results shown in Table 1 and Table 2, the optimum noise removal algorithm (the optimum combination of a plurality of noise removal algorithms and the optimum variable coefficient) depends on the location in the ship 1, the state of the engine, and the environmental conditions. Are different.
Table 3 below shows a correspondence example between the position of the speaker 8 and the engine speed in the ship 1 and the noise removal algorithm.

The noise removal algorithm switching unit 78 includes, for example, a memory 78A (see FIG. 4) that stores table information as shown in Table 3. In this memory 78A, experiments in various situations are performed, and identification information of an optimum noise removal algorithm is stored in advance corresponding to each situation.
In the case of the example in Table 3, the noise removal algorithm switching unit 78 is based on the position identification information provided from the IC tag reader 46 and the engine speed received from the ship 1 side via the communication processing unit 76. The identification information of the corresponding denoising algorithm is specified from the table in the memory. A noise removal algorithm (a set of coefficients γ, δ, and ε) corresponding to the identified identification information is read from the noise removal algorithm buffer 77 and set in the noise removal processing unit 73. That is, if the position of the speaker 8 in the ship 1 and the engine speed are uniquely determined, one noise removal algorithm optimal for the situation is selected from the noise removal algorithm buffer 77.

Of course, when the noise removal algorithm is switched using the attribute information and / or the environment information of the ship 1 in addition to the engine speed, table information for realizing the switching of the noise removal algorithm corresponding to them is prepared. .
As shown in FIG. 6, the preprocessing unit 89 framing the digital audio signal processed by the high-frequency emphasis filter unit 81 into frames, and the framing unit 85 framed them. And a frequency analysis unit 87 for performing frequency analysis processing of the audio signal. Similarly, the pre-processing unit 90 framing the digital audio signal processed by the high frequency emphasis filter unit 82, respectively, and frequency analysis processing of the audio signal framed by the framing unit 86. And a frequency analysis unit 88 for performing the operation.

The A / D converters 71 and 72 (see FIG. 4) are two-channel audio signals x ₁ (t) and x ₂ (t) from the first and second microphones 41 and 42 (t represents time). Are sampled at predetermined time intervals Δt. For example, signals y ₁ (i) and y ₂ (i) obtained by subjecting the sampled audio signals x ₁ (i) and x ₂ (i) to high-frequency emphasis filter processing are converted into a framing unit 85. , 86. The framing units 85 and 86 frame the audio signals y ₁ (i) and y ₂ (i) of a predetermined number of samples one after another.

The frequency analyzers 87 and 88 perform frequency analysis on the audio signals y ₁ (i) and y ₂ (i) by Fast Fourier Transform (FFT) in units of frames to obtain frequency functions X ₁ (ω), X ₂ (ω) is generated and passed to the spectral subtraction filter units 83 and 84.
FIG. 7 is a block diagram for explaining a configuration example of the utterance section detection unit 91. In this embodiment, the utterance section detection unit 91 uses the technique disclosed in Non-Patent Document 1. That is, the speech section is detected based on the cross spectrum of the two-channel audio signals y ₁ (i) and y ₂ (i) that have undergone the high-frequency emphasis filter processing.

Assume that the audio signal y ₂ (i) corresponding to the sound received by the second microphone 42 is a time movement waveform of the audio signal y ₁ (i) corresponding to the sound received by the first microphone 41. Then, if the delay time between the _two audio signals y ₁ (i) and y ₂ (i) is i ₀ · Δt, the audio signal y ₂ (i) can be expressed by the following equation (5).
y ₂ (i) = y ₁ (i−i ₀ ) (5)
In this case, the relationship between the frequency functions S ₁ (ω) and S ₂ (ω) is expressed by the following equation (6).

  On the other hand, when a change with respect to the frequency of the phase of the cross spectrum G (ω) is plotted on a two-dimensional plane for a certain frame in the non-speech interval, for example, the result is as shown in FIG. That is, in the non-utterance period, the phase component of the cross spectrum G (ω) does not have a constant trend with respect to the frequency. This is because the phase of the noise received by the first and second microphones 41 and 42 in the non-speech section is random.

Therefore, by examining whether the slope of the phase component of the cross spectrum G (ω) with respect to the frequency has a constant tendency, it is possible to discriminate between the frames in the speech zone and the frames in the non-speech zone.
In order to execute such processing, the utterance section detection unit 91 includes a cross spectrum calculation unit 100, a phase component extraction unit 101, a phase unwrap processing unit 102, and a main calculation unit 103. .

The cross spectrum calculation unit 100 performs an operation according to the expression (7) on the frequency functions S ₁ (ω) and S ₂ (ω) given from the noise removal processing unit 73 (spectral subtraction filter units 83 and 84). To obtain a cross spectrum G (ω). The phase component extraction unit 101 detects (extracts) the phase of the cross spectrum G (ω) obtained by the cross spectrum calculation unit 100. This detection result is given to the phase unwrap processing unit 102. The phase unwrap processing unit 102 performs unwrap processing for smoothing the discontinuous portion of the phase on the cross spectrum G (ω) based on the detected phase.

  The main calculation unit 103 is provided with a frequency band division unit 105 to which the cross spectrum G (ω) subjected to the unwrap processing from the phase unwrap processing unit 102 is provided, and N frequency segments generated by the frequency band division unit 105 are provided. 1st to Nth slope calculation units 106-1 to 106-N (hereinafter, collectively referred to as "first to Nth slope calculation units 106") and slopes obtained by the first to Nth slope calculation units 106, respectively. A histogram calculation unit 107 that generates a histogram related to the frequency of Further, the speech segment detection unit 91 further includes a speech segment determination unit 108 that determines whether each frame belongs to a speech segment or a non-speech segment based on the histogram generated by the histogram calculation unit 107, and an audio signal input. An on / off control unit 109 is provided.

The utterance section detection unit 91 further performs inverse Fourier transform to one channel frequency function S ₁ (ω) of the two channel frequency functions S ₁ (ω) and S ₂ (ω) given from the noise removal processing unit 73. And an inverse Fourier transform unit 104 that generates an audio signal z (i). The audio signal z (i) generated by the inverse Fourier transform unit 104 is supplied to the audio signal input on / off control unit 109.

  The frequency band division unit 105 divides (band division) the phase component of the cross spectrum G (ω) into small frequency segments (N segments). The first to Nth slope calculation units 106 calculate the slopes of the phase components with respect to the frequency, for example, using the least square method for the N frequency segments divided by the frequency band division unit 105. In order to obtain the inclination for each segment by the method of least squares, for example, a known technique as described in Non-Patent Document 2 can be applied.

  FIG. 9A is an example of a histogram obtained by the histogram calculation unit 107 for a frame in an utterance interval, and FIG. 9B is an example of a histogram obtained by the histogram calculation unit 107 for a frame in a non-utterance interval. These are histograms for the slopes obtained for each segment. That is, FIGS. 9 (a) and 9 (b) show the distribution of the slope of the phase, and the ratio of the number of segments of each slope (that is, the normalized frequency) to the total number of segments is a percentage. It is represented by

As is clear from the comparison between FIG. 9A and FIG. 9B, a clear peak value appears in the histogram in the speech interval frame. That is, the phase gradient is localized in a very narrow range. On the other hand, in the non-speech section frame, the histogram has a smooth shape, no clear peak, and the phase gradient is distributed over a wide range.
For example, the utterance section determination unit 108 determines whether each frame is an utterance section frame or a non-speech section frame based on the frequency of inclination of each phase in the histogram obtained by the histogram calculation section 107. That is, for example, when the appearance frequency of the phase inclination within the predetermined range p ± α centered on the theoretical value p of the peak of the phase inclination is equal to or higher than the predetermined threshold value β (%), the frame is included in the speech section. If the appearance frequency is less than the threshold value, the frame is determined to belong to a non-speech segment. The theoretical value p is “0” when the distances from the sound source (speech position) to the first and second microphones 41 and 42 are equal.

Α and β are examples of utterance detection parameters applied to the utterance section detection process. These parameters can be variably set, and a plurality of kinds of values are set by the utterance detection parameter switching unit 94 (see FIG. 5). Can be switched to.
The voice signal input on / off control unit 109 switches between an on state and an off state in response to the determination result of the utterance section determination unit 108. The on state is a state in which the frame-unit audio signal z (i) generated by the inverse Fourier transform unit 104 is passed to the acoustic analysis unit 92 (see FIG. 5). The off state is a state in which the frame unit audio signal z (i) generated by the inverse Fourier transform unit 104 is prevented from passing to the acoustic analysis unit 92. The speech signal input on / off control unit 109 is turned on when the speech segment determination unit 108 determines that the speech signal frame belongs to the speech segment, and the speech signal determination unit 108 determines that the speech signal frame belongs to the non-speech segment. When judged, it is turned off. Therefore, only the audio signal frame in the utterance section is supplied to the acoustic analysis unit 92.

If the acoustic analysis unit 92 performs fast Fourier transform (FFT) with the same specifications as the frequency analysis unit 87 and the frequency analysis unit 88, the frequency function S ₁ (ω) is used instead of the audio signal z (i). May be provided to the acoustic analysis unit 92 via the audio signal input on / off control unit 109. In this case, the inverse Fourier transform unit 104 is not necessary.
Table 4 below shows experimental results for obtaining optimum parameters α and β for a plurality of values of the engine speed at a plurality of locations in the ship 1.

From Table 4, it is understood that the optimum parameters α and β differ depending on the position in the ship 1 and the engine speed. Therefore, it is expected that optimum parameters α and β exist depending on the attributes of the ship 1 and the environment in which the ship 1 is placed.
Other methods for discriminating speech and noise can be applied to the speech segment detection processing. For example, a method is known in which speech and noise are discriminated based on the number of zero crossings (the number of times a signal waveform of unit time crosses zero) and the amplitude level of the waveform. More specifically, a threshold for the amplitude level of the input waveform is determined in advance according to the number of zero crossings per unit time. Then, the number of zero crossings and the amplitude level of the input waveform are detected, and when the amplitude level exceeds a threshold corresponding to the number of zero crossings, it is determined that the input signal is a signal in the utterance section. Also in this method, the utterance section detection accuracy can be improved by adjusting the optimum threshold according to the noise environment. That is, by setting the threshold value as an utterance detection parameter variably according to the position of the speaker 8 in the ship 1, the engine speed, and the like, highly accurate utterance section detection processing can be performed.

  Table 5 below shows a correspondence example between the position of the speaker 8 and the engine speed in the ship 1 and the speech detection parameter.

The speech detection parameter switching unit 94 includes, for example, a memory 94A (see FIG. 5) that stores table information as shown in Table 5. In the memory 94A, experiments in various situations are performed, and the identification information of the optimum speech detection parameter in the situation is stored in advance corresponding to each situation.
In the case of the example of Table 5, the speech detection parameter switching unit 94 is based on the position identification information provided from the IC tag reader 46 and the engine speed received from the ship 1 side via the communication processing unit 76. The identification information of the corresponding speech detection parameter is specified from the table in the memory 94A. The utterance detection parameter (the optimum set of parameters α and β) corresponding to the identified identification information is read from the utterance detection parameter buffer 95 and set in the utterance section detection unit 91.

Of course, when the utterance detection parameters are switched using the attribute information and / or environment information of the ship 1 in addition to the engine speed, table information for realizing the switching of the utterance detection parameters in accordance with them is prepared. .
Next, the principle of speech recognition will be described.
The acoustic analysis unit 92 analyzes the speech signal of the frame in the utterance section, and generates a feature vector x representing an acoustic feature. This feature vector x is collated by the collation unit 93 with the acoustic model and the recognition dictionary.

The acoustic model is obtained by modeling the acoustic features of speech, and is reference information for evaluating similarity to a partial or entire feature amount sequence of spoken speech. For example, the probability p (x | w) that the feature x is observed from the word (or phoneme) w to be recognized is obtained from the acoustic model.
The recognition dictionary (word dictionary or language model) is information for giving a restriction on connection of an acoustic model. A typical example of such information is a probability that another word appears after a certain word. For example, the probability p (w) that the word (or phoneme) w appears is obtained from the recognition dictionary.

The matching unit 93 obtains a probable word (or phoneme) w ′ for the feature vector x by the following equation (8) and outputs it as a recognition result.
w ′ = arg max _w p (x | w) p (w) (8)
In order to improve the speech recognition rate, it is preferable to create an acoustic model in a noise environment that actually uses speech recognition.

Therefore, in this embodiment, a plurality of acoustic models 931 to 93n corresponding to the attribute information and the environment information of the ship 1 are created in advance for the plurality of positions of the speaker 8 in the ship 1 and stored in the acoustic model buffer 97. Stored.
Table 6 below shows a correspondence example between the position of the speaker 8 and the engine speed in the ship 1 and the acoustic model.

The acoustic model switching unit 96 includes, for example, a memory 96A (see FIG. 5) that stores table information as shown in Table 6. The memory 96A stores in advance the identification information of the acoustic model that is optimal in the situation corresponding to each situation.
In the case of the example of Table 6, the acoustic model switching unit 96 is based on the position identification information provided from the IC tag reader 46 and the engine speed received from the ship 1 side via the communication processing unit 76. The identification information of the corresponding acoustic model is specified from the table in the memory 96A. The acoustic model corresponding to the identified identification information is read from the acoustic model buffer 97 and applied to the matching process in the matching unit 93.

Of course, when the acoustic model is switched using the attribute information and / or the environment information of the ship 1 in addition to the engine speed, table information for realizing the switching of the acoustic model according to them is prepared.
Next, the operation of the ship operation data generation unit 75 (see FIG. 4) to which the recognition result data is transferred will be outlined.

When the recognition result data is transferred from the voice recognition processing unit 74, the ship operation data generating unit 75 generates ship operation data corresponding to the ship operation command and transfers it to the communication processing unit 76 when it is a ship operation command. .
The marine vessel operation command includes, for example, a command for marine vessel maneuvering operation (marine maneuvering command) and a command for manipulating auxiliary equipment (attachment equipment maneuvering command).

Examples of ship maneuvering commands are as follows:
“Slowly forward”: Command to advance the vessel at a slow speed “Front”… Command to advance the vessel at low speed “Back”… Command to move the vessel backward “Tarely”… Stop (stop) the vessel Command “Small right” ・ ・ ・ Command to steer to the right at the first steering angle “Right” ・ ・ ・ Command to steer to the right at the second steering angle (> 1st steering angle) “Large right” ..Command to steer to the right at the third steering angle (> 2nd steering angle) "Small left" ... Command to steer to the left at the fourth steering angle "Left" ... Fifth steering angle Command to steer left at (> 4th steering angle) "Large left" ... Command to steer left at 6th steering angle (> 5th steering angle) For example, 1st and 4th steering angles Is one quarter of the maximum steering angle, the second and fifth steering angles are one half of the maximum steering angle, The third and sixth steering angles are maximum steering angles.

Examples of accessory device operation commands are as follows.
"Signal red flame" ... command to ignite signal red flame "Rubber boat" ... command to inflate the rubber boat and drop it onto the water "ship phone" ... command to start a phone call Ship operation data representing an operation command is transferred to the ship 1 side via the communication processing unit 76 and the communication device 44. On the ship 1 side, ship operation data is received by the wireless communication unit 16, and this ship operation data is transferred to the processing / output unit 17. The processing / output unit 17 operates the corresponding device in the ship 1 based on the received ship operation data. In this way, the ship operation can be performed by voice using the voice instruction device 10.

FIG. 10 is an illustrative view for explaining the configuration of a ship system according to the second embodiment of the present invention. In FIG. 10, functional parts equivalent to those shown in FIGS. 3 and 4 are given the same reference numerals.
In this embodiment, an arithmetic processing unit 125 for voice recognition processing and ship operation data generation is provided on the ship 1A side, and this arithmetic processing unit 125 is connected to the inboard LAN 11. The arithmetic processing unit 125 includes a noise removal processing unit 73, a voice recognition processing unit 74, a ship operation data generation unit 75, a noise removal algorithm buffer 77, and a noise removal algorithm switching unit 78, and for communication with the inboard LAN 11. The communication interface unit 126 is provided.

On the other hand, the arithmetic processing unit 45A on the voice instruction device 10A side includes A / D conversion units 71 and 72 and a communication processing unit 76. A / D conversion units 71 and 72 and an IC tag reading device 46 are connected to the communication processing unit 76.
The communication processing unit 76 performs processing for wirelessly transmitting the audio signal input from the microphones 41 and 42 and converted into a digital signal by the A / D conversion units 71 and 72 to the ship 1A side via the communication device 44. Execute. In addition, the communication processing unit 76 executes processing for wirelessly transmitting the position identification information read by the IC tag reading device 46 to the ship 1 </ b> A side via the communication device 44.

  The arithmetic processing unit 125 of the ship 1A receives a voice signal from the voice instruction device 10A side via the wireless communication unit 16 and the inboard LAN 11, and executes voice recognition processing on the voice signal. Further, the arithmetic processing unit 125 obtains the position identification information from the voice instruction device 10 </ b> A in the same manner as the voice signal, and gives it to the noise removal algorithm switching unit 78 and the voice recognition processing unit 74. Further, the arithmetic processing unit 125 acquires the attribute information and environment information of the ship 1 </ b> A via the communication interface unit 126, and provides them to the noise removal algorithm switching unit 78 and the voice recognition processing unit 74.

Even with such a configuration, the speaker 8 can give an instruction for ship operation by voice. Since the parameters of the noise removal process and the voice recognition process are switched according to the position of the speaker 8 in the ship 1A and the attribute information and environment information of the ship 1, a high voice recognition rate can be obtained.
While the two embodiments of the present invention have been described above, the present invention can also be implemented in other forms. For example, in the above-described embodiment, the audio signal given from the speech section detection unit 91 to the acoustic analysis unit 92 is a one-channel audio signal corresponding to the first microphone 41, and an acoustic signal is generated based on the one-channel audio signal. Analysis processing is performed. However, in order to further increase the voice recognition rate, it is preferable to perform acoustic analysis processing using the voice signals of both channels of the first and second microphones 41 and 42.

In order to use such channel-synchronized addition speech, for example, in the configuration of FIG. 7, the frequency functions S ₁ (ω) and S ₂ (ω) are converted into the acoustic analysis unit 92 via the speech signal input on / off control unit 109. To give. Then, the acoustic analysis unit 92 may obtain the channel-synchronized addition sound as described above, and perform an acoustic analysis on the channel-synchronized addition sound.
In the above-described embodiment, the example in which the noise removal algorithm, the speech detection parameter, and the acoustic model are switched according to the position of the speaker 8 in the ship 1 is described. Of these three parameters, It is good also as a structure which switches only 1 type or 2 types of these. For example, only the noise removal algorithm may be switched, and the parameters for speech recognition processing may be fixed. Further, the noise removal algorithm may be fixed and the parameters of the speech recognition process may be changed. Furthermore, in the speech recognition process, either one of the speech detection parameter and the acoustic model may be fixed and the other may be changed.

Furthermore, a noise removal apparatus that performs noise removal processing can be configured as voice signal processing that does not involve voice recognition processing. In this case, the configuration of FIG. 4 may be configured such that the voice recognition processing unit 74 and the ship operation data generation unit 75 are omitted.
In the configuration of the first embodiment, the ship operation data generation unit 75 may be provided on the ship 1 side, and the recognition result data may be transmitted to the ship 1 side via the communication processing unit 76.

  Furthermore, in the first and second embodiments described above, the position of the speaker 8 can be specified by reading the position identification information of the IC tags 61 to 6M arranged at a plurality of locations on the ship 1 with the IC tag reader 46. However, the position of the speaker 8 can be specified by other methods. For example, the voice instruction device 10 may be provided with a key input unit, and the speaker 8 may input the position identification information from the key input unit. In addition, for example, a transmission device that transmits position identification information in response to a user's operation (for example, transmission by an optical signal) is provided at a plurality of locations on the ship 1, and the position identification information transmitted from the transmission device is provided. May be specified by the voice instruction device 10 to identify the position of the speaker 8. In this case, the transmitting device serves as an identification information carrier, and the receiving device provided on the voice instruction device 10 serves as a reading device that reads position identification information.

  In the first and second embodiments described above, information is exchanged between the voice instruction device 10 and the inboard LAN 11 by wireless communication, but communication between these is wired using a communication cable. It may be performed by communication. In this case, connection outlets may be provided at a plurality of locations on the ship 1, and the voice instruction device 10 may be connected to any one of the connection outlets. Further, in this case, it may be determined to which connection outlet the voice instruction device 10 is connected, and thereby the position of the speaker 8 may be specified.

  It is also conceivable to modify the configuration of the second embodiment described above so that a plurality of microphones for inputting the voice of the speaker 8 are arranged at a plurality of locations in the ship 1. In this case, the speaker 8 holds an IC tag carrying predetermined authentication information, and places a reading device in the vicinity of each microphone installation location. In use, the speaker 8 holds the IC tag over the reading device and reads the authentication information recorded on the IC tag. The read authentication information is sent to the arithmetic processing unit 125 via the inboard LAN 11, and the position of the speaker 8 is specified depending on which location the authentication information is input by the reading device. In the arithmetic processing unit 125, the algorithm applied to the noise removal processing and the parameters applied to the speech recognition processing are switched based on the specified position information.

Further, in the above-described embodiment, the configuration in which the attribute information and the environment information of the ship 1 are taken into the voice instruction device 10 by communication with the inboard LAN 11 is described. For example, a key input unit is provided in the voice instruction device 10. It is also possible to provide attribute information and environment information by operating the key input unit.
In the above-described embodiment, the environmental information is detected by the environmental information sensor 9. For example, the wireless communication unit 16 communicates with one of the information centers and wind speed in the vicinity of the ship 1. It is good also as a structure which acquires environmental information, such as a wave height.

Moreover, in the above-described embodiment, a ship is taken as an example of the transportation device. However, the present invention can be suitably applied to a large transportation device in which an occupant can move inside, such as a train or an aircraft.
Furthermore, in the above-described embodiment, the case where the voice of the speaker 8 is set as the processing target sound has been described. The present invention may be applied to a device that performs self-diagnosis of transport equipment based on the mechanical sound of

  In the above-described embodiment, the example in which the voice instruction device 10 is formed in a transceiver shape and the microphones 41 and 42 are attached to the casing has been described. However, in order to keep the positional relationship between the utterance position (mouth) of the speaker 8 and the microphones 41 and 42 constant and improve the speech recognition rate, it is preferable to use a headset type microphone. Further, the microphone may be attached to a life jacket that is required to be worn by a crew member of the ship or recommended to be worn. In this case, two microphones may be attached to the left and right shoulders of the life jacket, or a stereo microphone may be used.

Of course, a two-channel sound reception system is not necessarily required, and depending on the type of speech segment detection processing, acoustic analysis processing, or the like, a one-channel sound reception system may be sufficient.
Further, in the above-described embodiment, the apparatus for processing the audio signal from the microphones 41 and 42 has been described. However, the present invention can process the audio signal obtained by reproducing the sound recorded in advance by the recording apparatus with the reproducing apparatus. It is also possible to apply to.

  In addition to the above, various design changes can be made within the scope of the matters described in the claims.

It is an illustration figure which shows the notional structure of the ship system as an example of the transport equipment system which concerns on one Embodiment of this invention. It is a schematic top view of the ship which comprises the said ship system. It is a block diagram which mainly shows the electrical structure inside a ship. It is a block diagram for demonstrating the electrical structure of a voice instruction | indication apparatus. It is a block diagram for demonstrating the functional structure of a speech recognition process part. It is a block diagram which shows the structural example of a noise removal process part. It is a block diagram for demonstrating the structural example of an utterance area detection part. (a) shows an example of the change of the cross spectrum phase with respect to the frequency of the frame in the utterance interval, and (b) shows an example of the change of the cross spectrum phase with respect to the frequency of the frame in the non-utterance interval. (a) is a histogram about the phase gradient obtained for the frame in the speech segment, and (b) is a histogram about the phase gradient obtained for the frame in the non-speech segment. It is an illustration figure for demonstrating the structure of the ship system which concerns on 2nd Embodiment of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1,1A Ship 2 Outboard motor 3 Steering device 4 Signal flame device 5 Ship telephone device 61-6M IC tag 7 Rubber boat operation device 8 Speaker 9 Environment information sensor 10, 10A Voice indication device 11 Inboard LAN
DESCRIPTION OF SYMBOLS 12 Database apparatus 13 Steering part 14 Shift / throttle operation part 16 Wireless communication part 17 Processing / output part 18 Steering handle 20 Sensors, such as an engine speed sensor 21 Engine 22 Engine control unit 41 1st microphone 42 2nd microphone 44 Communication apparatus 45, 45A arithmetic processing unit 46 IC tag reader 51 hull 52 superstructure 53 out deck 54 cabin 55 cockpit 56 flying deck 73 noise removal processing unit 731-73k noise removal algorithm 74 voice recognition processing unit 75 ship operation data generation unit 76 Communication processing unit 77 Noise removal algorithm buffer 78 Noise removal algorithm switching unit 81, 82 High frequency enhancement filter unit 83, 84 Spectral subtraction filter unit 85, 8 6 Framing unit 87, 88 Frequency analysis unit 89, 90 Pre-processing unit 91 Speech segment detection unit 911-91m Speech segment detection parameter 92 Acoustic analysis unit 93 Collation unit 931-93n Acoustic model 94 Speech detection parameter switching unit 95 Speech detection parameter Buffer 96 Acoustic model switching unit 97 Acoustic model buffer 98 Recognition dictionary 100 Cross spectrum calculation unit 107 Histogram calculation unit 108 Speech segment determination unit 109 Voice signal input on / off control unit 125 Arithmetic processing unit 126 Communication interface unit

Claims (14)

A target sound processing device for processing a predetermined target sound generated in a transport device,
A generation position specifying means for specifying a generation position of the target sound in the transport device;
A sound signal processing means that receives a sound signal that is an electrical signal of a sound including the target sound, executes sound signal processing for processing the sound signal, and can change parameters of the sound signal processing;
A target sound processing apparatus comprising: parameter switching means for switching a parameter of the sound signal processing in the sound signal processing means in accordance with the generation position specified by the generation position specifying means. A transport device information acquiring means for acquiring transport device information including at least one of attribute information of the transport device and environmental information of the transport device;
The parameter switching means is a parameter of the sound signal processing in the sound signal processing means based on the transport equipment information acquired by the transport equipment information acquisition means in addition to the generation position specified by the generation position specifying means. The target sound processing device according to claim 1, wherein the target sound processing device is switched. The transport device includes a database storing attribute information of the transport device,
The target sound processing apparatus according to claim 2, wherein the transport device information acquisition unit includes attribute information acquisition unit that acquires attribute information of the transport device from the database. The transport equipment includes transport equipment information detection means for detecting the transport equipment information,
4. The target sound processing apparatus according to claim 2, wherein the transport device information acquisition unit acquires the transport device information detected by the transport device information detection unit from the transport device. The sound signal processing means includes noise removal processing means for performing noise removal processing for removing a noise component from the input sound signal,
5. The object according to claim 1, wherein the parameter switching unit includes a noise removal parameter switching unit that switches a noise removal parameter applied to a noise removal process performed by the noise removal processing unit. Sound processing device. The sound signal processing means includes target sound extraction means for performing target sound section extraction processing for extracting a signal section including the target sound from the input sound signal,
6. The parameter switching means includes a target sound extraction parameter switching means for switching a target sound extraction parameter applied to a target sound section extraction process executed by the target sound extraction means. The target sound processing device according to 1. The sound signal processing means has a plurality of types of acoustic models that can be collated with the sound signal of the target sound,
7. The target sound processing apparatus according to claim 1, wherein the parameter switching means includes acoustic model switching means for switching the plurality of types of acoustic models. The transport device is disposed at a plurality of different positions in the transport device, and includes a plurality of identification information carriers each carrying position identification information,
The target sound processing apparatus according to claim 1, wherein the generation position specifying unit includes a reading device that reads position identification information carried on the identification information carrier. The transport equipment is a ship;
9. The target sound processing apparatus according to claim 1, wherein the generation position specifying unit specifies a generation position of the target sound in the ship. The target sound is a voice uttered by a speaker in the transport device,
The target sound processing apparatus according to claim 1, wherein the sound signal processing means includes voice recognition processing means for executing voice recognition processing for recognizing the voice of the speaker. Transportation equipment,
A transportation equipment system comprising the target sound processing device according to claim 1. Transportation equipment,
The target sound processing device according to claim 10;
Based on the voice recognition result by the voice recognition processing unit, the instruction for operation of the transport device by the speaker is specified, and transport device operation data generating unit for generating transport device operation data corresponding to the instruction
A transport equipment system comprising: equipment control means for operating equipment provided in the transport equipment based on the transport equipment operation data generated by the transport equipment operation data generation means. A target sound processing method for processing a predetermined target sound generated in a transport device,
Identifying the location of the target sound in the transport device;
Performing sound signal processing on a sound signal that is an electrical signal of a sound including the target sound;
And a step of switching a parameter applied to the sound signal processing in accordance with the identified position where the target sound is generated. Further comprising obtaining transport equipment information including at least one of attribute information of the transport equipment and environmental information of the transport equipment,
The step of switching the parameter includes a step of switching a parameter applied to the sound signal processing based on the acquired transport device information in addition to the identified generation position of the target sound. The target sound processing method according to claim 13.

Images