WO2019113911A1 - 设备控制方法、云端设备、智能设备、计算机介质及设备 - Google Patents

设备控制方法、云端设备、智能设备、计算机介质及设备 Download PDF

Info

Publication number
WO2019113911A1
WO2019113911A1 PCT/CN2017/116319 CN2017116319W WO2019113911A1 WO 2019113911 A1 WO2019113911 A1 WO 2019113911A1 CN 2017116319 W CN2017116319 W CN 2017116319W WO 2019113911 A1 WO2019113911 A1 WO 2019113911A1
Authority
WO
WIPO (PCT)
Prior art keywords
wake
word
same
words
smart
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2017/116319
Other languages
English (en)
French (fr)
Inventor
邹莹
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Haier Uplus Intelligent Technology Beijing Co Ltd
Original Assignee
Haier Uplus Intelligent Technology Beijing Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Haier Uplus Intelligent Technology Beijing Co Ltd filed Critical Haier Uplus Intelligent Technology Beijing Co Ltd
Priority to PCT/CN2017/116319 priority Critical patent/WO2019113911A1/zh
Priority to CN201780097241.4A priority patent/CN111386566A/zh
Publication of WO2019113911A1 publication Critical patent/WO2019113911A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00—Speech recognition
    • G10L15/06—Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00—Speech recognition
    • G10L15/22—Procedures used during a speech recognition process, e.g. man-machine dialogue
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00—Speech recognition
    • G10L15/26—Speech to text systems
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
    • G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation

Definitions

  • the embodiments of the present invention relate to the field of Internet technologies, and in particular, to a device control method, a cloud device, a smart device, a computer medium, and a device.
  • the existing smart home devices generally support the voice wake-up function. For example, the user sends a wake-up word, and after the smart home device uses the voice collection device to collect the voice signal, the content of the voice signal is recognized as the wake-up word set by the smart home device. , you can automatically turn on this smart home device.
  • the increase of smart home devices in the home it is possible that multiple smart home devices use the same wake-up word, which is easy for the user to accurately control the device that they want to control, causing trouble for the user and affecting the effective command. carried out.
  • the present invention provides a device control method, a cloud device, a smart device, a computer medium, and a device.
  • the device control method provided in this article is applied to cloud devices, including:
  • the above device control method also has the following features:
  • the method further includes: determining the same wake-up word according to the wake-up word update rule.
  • the above device control method also has the following features:
  • the wake-up word update rule includes adding a suffix with a sequential relationship to the wake-up word.
  • the above device control method also has the following features:
  • the method further includes:
  • an instruction to modify the same wake-up word is generated, including:
  • a computer readable storage medium is provided, on which a computer program is stored, the steps of which are implemented when the program is executed by the processor.
  • a computer device comprising a memory, a processor, and a computer program stored on the memory and operable on the processor, the processor implementing the steps of the method.
  • the cloud devices provided in this article include:
  • a first acquiring module configured to acquire wake words of multiple smart devices in the same voice receiving area, where multiple smart devices can be woken up by respective wake words
  • a first detecting module configured to detect whether the same wake-up word is included in the wake-up words of the multiple smart devices
  • the first generating module is configured to generate an instruction to modify the same wake-up word if the same wake-up word is included in the wake-up words of the plurality of smart devices.
  • a second obtaining module configured to acquire, when the first obtaining module acquires the wake-up words of the plurality of smart devices in the same voice receiving area, the hardware addresses of the routers of the plurality of smart devices that are in the same voice receiving area;
  • the first detecting module includes a second detecting module, configured to detect, according to the hardware address of the router to which the plurality of smart devices belong, whether there are smart devices belonging to the same router among the plurality of smart devices, and multiple intelligent devices If there is a smart device belonging to the same router in the standby, it is detected whether there is the same wake-up word in the wake-up word of the smart device belonging to the same router;
  • the first generation module includes a second generation module, configured to generate an instruction to modify the same wake-up word if the wake-up words of the smart device belonging to the same router have the same wake-up word.
  • the device control method provided in this paper is applied to smart devices, including:
  • the cloud device When the cloud device detects that the wake-up words of other smart devices in the same voice receiving area as the smart device are the same as the wake-up words of the smart device, receiving and storing the new wake-up word sent by the cloud device or receiving the modified smart device sent by the cloud device The instruction to wake up the word.
  • the above device control method also has the following features:
  • the method After receiving the instruction sent by the cloud device to modify the wake-up word of the smart device, the method further includes: determining a new wake-up word according to the wake-up word update rule or receiving and storing a new wake-up word set by the user according to the instruction.
  • the above device control method also has the following features:
  • the method further includes: changing the pre-built speech recognition decoding network according to the new wake-up word.
  • the above device control method also has the following features:
  • the pre-constructed acoustic model is a model obtained by training a full phoneme, a garbage phoneme and a noise phoneme using the first criterion and the second criterion
  • the first criterion is a maximum likelihood estimation criterion
  • the second criterion is a minimum phoneme error criterion Differentiated training criteria for optimization criteria
  • the sequence of state index values corresponding to the original wake-up words in the pre-constructed speech recognition decoding network is replaced with the sequence of state index values corresponding to the new wake-up words.
  • the above device control method also has the following features:
  • the method further includes:
  • the above device control method also has the following features:
  • Sending the wake-up word of the smart device to the cloud device the method further includes: sending, to the cloud device, a hardware address of the router to which the smart device belongs;
  • the cloud device When the cloud device detects that the wake-up words of other smart devices in the same voice receiving area as the smart device are the same as the wake-up words of the smart device, receiving and storing the new wake-up word sent by the cloud device or receiving the modified smart device sent by the cloud device Instructions for awakening words, including:
  • Receiving and storing the new wake-up word or receiving the cloud device sent by the cloud device when the cloud device detects that the wake-up word of the other smart device in the same voice receiving area as the smart device and the hardware address of the router is the same as the wake-up word of the smart device An instruction to modify the wake-up word of the smart device.
  • Another computer readable storage medium provided herein, the computer program having a computer program stored thereon, the steps of the method being implemented when the program is executed by the processor.
  • Another computer device includes a memory, a processor, and a computer program stored on the memory and operable on the processor, the processor implementing the steps of the method.
  • the smart devices provided in this article include:
  • a first sending module configured to send, to the cloud device, a wake-up word of the smart device, where the smart device can be woken up by the wake-up word;
  • the first receiving module is configured to: when the cloud device detects that the wake-up words of other smart devices in the same voice receiving area as the smart device are the same as the wake-up words of the smart device, receive the new wake-up word sent by the cloud device or send the cloud device to send The instruction to modify the wake-up word of the smart device.
  • Smart devices also include:
  • the second receiving module is configured to receive and store a new wake-up word set by the user according to the instruction after the first receiving module receives the instruction for modifying the wake-up word of the smart device sent by the cloud device;
  • the wake-up word update module is configured to determine a new wake-up word according to the wake-up word update rule after the first receiving module receives the instruction sent by the cloud device to modify the wake-up word of the smart device.
  • the smart device also includes a network update module for updating the new wake-up word according to the new wake-up word Change the pre-built speech recognition decoding network;
  • the network update module includes a calculation module and a network maintenance module
  • a calculation module configured to determine a sequence of triphone nodes corresponding to the new wake-up word, and determine a state index value corresponding to each three phoneme in the sequence of the triphone node according to the correspondence between the triphone and the model output state index value in the pre-built acoustic model, Forming a sequence of state index values;
  • the network maintenance module is configured to replace the sequence of state index values corresponding to the original wake-up words in the pre-constructed voice recognition decoding network with the sequence of state index values corresponding to the new wake-up words.
  • the smart device further includes a training module for training the full phoneme, the garbage phoneme and the noise phoneme to obtain a pre-constructed acoustic model by using the first criterion and the second criterion, the first criterion is a maximum likelihood estimation criterion, and the second criterion is a minimum phoneme
  • the error criterion is a differentiated training criterion for the optimization criterion.
  • Smart devices also include:
  • a second sending module configured to: when the first sending module sends the wake-up word of the smart device to the cloud device, send the hardware address of the router to which the smart device belongs to the cloud device;
  • the first receiving module includes a third receiving module, configured to receive and receive, when the cloud device detects that the wake-up words of other smart devices that are in the same voice receiving area as the smart device and the hardware address of the router are the same as the wake-up words of the smart device. Stores a new wake-up word sent by the cloud device or receives an instruction sent by the cloud device to modify the wake-up word of the smart device.
  • This document can effectively identify multiple smart devices in the same voice receiving area that use the same wake-up word, and generate an instruction to modify the same wake-up word.
  • the smart device can modify or notify the smart device to modify the wake-up word to prevent the same between smart devices.
  • the situation of wake-up words ensures the user's accurate control of the smart device and improves the user experience.
  • Embodiment 1 is a flowchart of a device control method applied to a cloud device in Embodiment 1;
  • FIG. 2 is a structural diagram of a cloud device in the first embodiment
  • Embodiment 3 is a flowchart of a device control method applied to a cloud device in Embodiment 2;
  • FIG. 4 is a structural diagram of a smart device in the second embodiment.
  • the smart device in the embodiment of the present invention is typically a smart home appliance.
  • a device control method applied to a cloud device includes:
  • Step 101 Acquire wake words of multiple smart devices in the same voice receiving area, where multiple smart devices can be woken up by respective wake words;
  • Step 102 Detect whether there are the same wake-up words in the wake-up words of the multiple smart devices
  • Step 103 In the case that the same wake-up word is included in the wake-up words of the plurality of smart devices, an instruction to modify the same wake-up word is generated.
  • the embodiment of the present invention can effectively identify multiple smart devices in the same voice receiving area that use the same wake-up word, and generate an instruction to modify the same wake-up word. According to the command, the smart device can be modified or notified to modify the wake-up word to prevent the smart device.
  • the same wake-up words appear between the users to ensure the user's accurate control of the smart device and improve the user experience.
  • the method can also include the step 104 of determining the same wake-up word based on the wake-up word update rule.
  • the wake-up word update rule includes adding a suffix word having a sequential relationship to the wake-up word. For example, if the original wake-up word is “power on”, the updated wake-up words are “power on 1”, “power on 2”, “power on 3”, etc.; or, the updated wake-up words are “boot A” and “boot B”. , "boot C" and so on.
  • the method body for updating the wake-up word in this embodiment may be a cloud device.
  • the embodiment of the present invention provides more than one type of update mode, so that the user can select a corresponding manner according to the use requirement, thereby improving the user experience.
  • the method further includes: acquiring the hardware addresses of the routers of the plurality of smart devices in the same voice receiving area.
  • step 102 detecting whether there are the same wake-up words in the wake-up words of the plurality of smart devices, including: detecting, by the hardware addresses of the routers to which the plurality of smart devices belong, whether the smart devices belonging to the same router are included in the plurality of smart devices; In the case where there are smart devices belonging to the same router in the smart device, it is detected whether there is the same wake-up word in the wake-up words of the smart devices belonging to the same router.
  • the instruction to modify the same wake-up word is generated in step 103, including: if the wake-up words of the smart devices belonging to the same router have the same wake-up word, Generate an instruction that modifies the same wake-up word.
  • smart devices that are in the same local area network and use the same wake-up words are used as smart devices in the same voice receiving area, considering that smart devices that are generally in the same local area network are generally located in the same small area (eg, home, company, etc.).
  • the distance between each smart device is small, it is easy to use the same wake-up word and multiple smart devices will receive this wake-up word when the user makes a voice wake-up word.
  • the first embodiment further includes a computer readable storage medium having a computer program stored thereon, and the steps of the foregoing method are implemented when the program is executed by the processor.
  • Embodiment 1 further includes a computer device including a memory, a processor, and a computer program stored on the memory and operable on the processor, the processor implementing the program to implement the steps of the foregoing method.
  • a computer device including a memory, a processor, and a computer program stored on the memory and operable on the processor, the processor implementing the program to implement the steps of the foregoing method.
  • the cloud device in the first embodiment includes:
  • a first acquiring module configured to acquire wake words of multiple smart devices in the same voice receiving area, where multiple smart devices can be woken up by respective wake words
  • a first detecting module configured to detect whether the same wake-up word is included in the wake-up words of the multiple smart devices
  • the first generating module is configured to generate an instruction to modify the same wake-up word if the same wake-up word is included in the wake-up words of the plurality of smart devices.
  • the cloud device further includes a second obtaining module, configured to acquire, when the first obtaining module acquires the wake-up words of the plurality of smart devices in the same voice receiving area, the hardware of the routers of the plurality of smart devices in the same voice receiving area address.
  • the first detecting module includes a second detecting module, configured to detect, according to the hardware address of the router to which the plurality of smart devices belong, whether there are smart devices belonging to the same router among the plurality of smart devices, and the smart devices belonging to the same router among the plurality of smart devices In the case of detecting whether there is the same wake-up word in the wake-up words of the smart devices belonging to the same router.
  • the first generation module includes a second generation module, configured to generate an instruction to modify the same wake-up word if the wake-up words of the smart device belonging to the same router have the same wake-up word.
  • the cloud device further includes: a wake-up word modification module, configured to determine the same wake-up word according to the wake-up word update rule.
  • the wake-up word update rule includes adding a suffix with a sequential relationship to the wake-up word. For example, if the original wake-up word is “power on”, the updated wake-up words are “power on 1”, “power on 2”, “power on 3”, etc.; or, the updated wake-up words are “boot A” and “boot B”. , "boot C" and so on.
  • a device control method applied to a smart device includes:
  • Step 301 Send a wake-up word of the smart device to the cloud device, where the smart device can be woken up by the wake-up word;
  • Step 302 When the cloud device detects that the wake-up words of other smart devices in the same voice receiving area as the smart device are the same as the wake-up words of the smart device, receive and store the new wake-up word sent by the cloud device or receive the modification sent by the cloud device. The instruction of the wake-up word of the smart device.
  • the method after receiving the instruction of the cloud device to modify the wake-up word of the smart device, the method further includes: determining a new wake-up word according to the wake-up word update rule or receiving and storing a new wake-up word set by the user according to the instruction.
  • the method body for updating the wake-up word may be a smart device or a user.
  • the embodiment of the present invention provides more than one. The update method allows the user to select the appropriate method according to the usage requirements and improve the user experience.
  • the smart device sends the wake-up words of the smart device to the cloud device, and further includes: sending the hardware of the router to which the smart device belongs to the cloud device. Address: when the cloud device detects that the wake-up words of other smart devices in the same voice receiving area as the smart device are the same as the wake-up words of the smart device, in step 302, the new wake-up word sent by the cloud device is received and stored, or the receiving cloud device sends the address.
  • the instruction for modifying the wake-up word of the smart device specifically includes: when the cloud device detects that the wake-up word of the other smart device in the same voice receiving area as the smart device and the hardware address of the router is the same as the wake-up word of the smart device, receiving and Stores a new wake-up word sent by the cloud device or receives an instruction sent by the cloud device to modify the wake-up word of the smart device.
  • the method further comprises: changing the pre-constructed speech recognition decoding network according to the new wake-up word.
  • the new wake-up word may be determined by receiving and storing a new wake-up word sent by the cloud device, or may be a new wake-up word determined by the smart device according to the wake-up word update rule or the smart device receiving and storing the user according to the instruction. Set a new wake up word.
  • the intelligent device pre-builds an acoustic model for implementing the wake-up word wake-up function
  • the pre-built acoustic model is a model obtained by training the full phoneme, the garbage phoneme and the noise phoneme using the first criterion and the second criterion.
  • the first criterion is a Maximum Likelihood Estimation (MLE) criterion
  • the second criterion is a differentiated training criterion based on a Minimum Phone Error (MPE) criterion.
  • MLE Maximum Likelihood Estimation
  • MPE Minimum Phone Error
  • the garbage model is trained according to the pronunciation variation law using the full phoneme in the full phoneme speech library to form a garbage phoneme library composed of M consonant phonemes and N vowel phonemes; because the garbage factor model is composed of multiple normals
  • the phoneme is mixed and trained, so the acoustic feature of the input speech when the wake-up word is encountered during decoding is scored on the junk phoneme, but the score on the phoneme corresponding to the wake-up word is competing, and when the speech other than the wake-up word is encountered, Because the phoneme contains the corresponding phoneme, it will naturally compete with the phoneme corresponding to the wake-up word.
  • this acoustic model increases the training part of the noise phoneme and can improve the anti-noise performance of the acoustic model.
  • the great advantage of constructing a full phoneme acoustic model is that it does not need to specify a specific wake-up word during training, and only a full likelihood method can be used to generate a full-coverage triphone hidden Markov model (Hidden). Markov Model, HMM).
  • the full phoneme acoustic model is trained using the full phoneme speech library.
  • the speech data of the wake-up speech range can be collected to establish a wake-up speech database, and the speech library is used to update.
  • the above acoustic model makes the above acoustic model have better matching to wake words.
  • the decoding device is also constructed by the smart device, and the decoding network includes a parallel garbage phoneme index value node, a noise phoneme index value node, and a sequence of state index values corresponding to the currently used wake-up word.
  • the decoding network is a recyclable network. The characteristics of this loop are that it can jump back from the exit node to the incoming node. You can cover multiple voice segments that continue.
  • the method for changing a pre-constructed speech recognition decoding network according to a new wake-up word comprises: determining a sequence of a triphone node corresponding to a new wake-up word, and determining a triphone node according to a correspondence between a triphone and a model output state index value in the pre-built acoustic model
  • the state index value corresponding to each triphone in the sequence constitutes a sequence of state index values; the sequence of state index values corresponding to the original wake-up words in the pre-constructed speech recognition decoding network is replaced with the sequence of state index values corresponding to the new wake-up words.
  • the manner of determining the sequence of the triphone nodes corresponding to the wake words is to use the phonemes of each word to form the triphones by combining the phonemes before and after.
  • the first phoneme of the first triphone and the last phoneme of the last triphone may also be included in the silencing phone sequence when acquiring the triphone node sequence.
  • the method for using the new wake-up word includes: receiving the voice data input by the user; extracting the voice feature from the voice data, for example, the voice feature is Meier The frequency-cepstrum coefficient (MFCC) decodes the speech feature using the speech recognition decoding network to obtain a decoding result; in the case where the decoding result includes a new wake-up word, the wake-up operation is performed.
  • MFCC frequency-cepstrum coefficient
  • the second embodiment further provides a computer readable storage medium, wherein the computer program is stored on the storage medium, and the steps of the foregoing method are implemented when the program is executed by the processor.
  • a second embodiment further provides a computer device comprising a memory, a processor, and a computer program stored on the memory and operable on the processor, the processor implementing the program to implement the steps of the method.
  • the smart device in the second embodiment includes:
  • a first sending module configured to send, to the cloud device, a wake-up word of the smart device, where the smart device can be woken up by the wake-up word;
  • a first receiving module configured to: when the cloud device detects that the wake-up words of other smart devices in the same voice receiving area as the smart device are the same as the wake-up words of the smart device, receive and store the new wake-up word or the receiving cloud sent by the cloud device The instruction sent by the device to modify the wake-up word of the smart device.
  • the smart device also includes a second receiving module or a wake-up word update module.
  • the second receiving module is configured to receive and coexist after the first receiving module receives the instruction for modifying the wake-up word of the smart device sent by the cloud device Store new wake-up words set by the user according to the instructions.
  • the wake-up word update module is configured to determine a new wake-up word according to the wake-up word update rule after the first receiving module receives the instruction sent by the cloud device to modify the wake-up word of the smart device.
  • the smart device also includes a network update module for modifying the pre-built speech recognition decoding network based on the new wake-up word after the new wake-up word is determined.
  • the network update module includes a calculation module and a network maintenance module.
  • the calculation module is configured to determine a sequence of a triphone node corresponding to the new wake-up word, and determine a state index value corresponding to each three phoneme in the sequence of the triphone node according to the correspondence between the triphone and the model output state index value in the pre-built acoustic model, and constitute a state index value.
  • a sequence of status index values are configured to determine a sequence of a triphone node corresponding to the new wake-up word, and determine a state index value corresponding to each three phoneme in the sequence of the triphone node according to the correspondence between the triphone and the model output state index value in the pre-built acoustic model, and constitute a state index value.
  • the network maintenance module is configured to replace the sequence of state index values corresponding to the original wake-up words in the pre-constructed voice recognition decoding network with the sequence of state index values corresponding to the new wake-up words.
  • the smart device further includes a training module for training the full phoneme, the garbage phoneme and the noise phoneme using the first criterion and the second criterion to obtain the above-prepared acoustic model, the first criterion is a maximum likelihood estimation criterion, and the second criterion is a minimum
  • the phoneme error criterion is a discriminative training criterion for the optimization criterion.
  • the smart device further includes a second sending module, configured to send, to the cloud device, a hardware address of the router to which the smart device belongs, while the first sending module sends the wake-up word of the smart device to the cloud device.
  • the first receiving module includes a third receiving module, configured to receive, when the cloud device detects that the wake-up words of other smart devices that are in the same voice receiving area as the smart device and the hardware address of the router are the same as the wake-up words of the smart device. And storing a new wake-up word sent by the cloud device or receiving an instruction sent by the cloud device to modify the wake-up word of the smart device.
  • the smart device further includes a fourth receiving module, an extracting module, a decoding module, and an executing module, wherein the fourth receiving module is configured to receive the voice data input by the user after the network update module changes the pre-built voice recognition decoding network according to the new wake-up word.
  • An extraction module configured to extract a speech feature from the speech data; a decoding module, configured to decode the speech feature using the speech recognition decoding network to obtain a decoding result; and an execution module, configured to: when the decoding result includes a new wake-up word , perform a wake-up operation.
  • computer storage medium includes volatile and nonvolatile, implemented in any method or technology for storing information, such as computer readable instructions, data structures, program modules or other data. Sex, removable and non-removable media.
  • Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridge, magnetic tape, magnetic disk storage or other magnetic storage device, or may Any other medium used to store the desired information and that can be accessed by the computer.
  • communication media typically includes computer readable instructions, data structures, program modules, or other data in a modulated data signal, such as a carrier wave or other transport mechanism, and can include any information delivery media. .
  • This paper can effectively identify multiple smart devices in the same voice receiving area using the same wake-up word, and actively modify or notify the smart device to modify the wake-up words to prevent the same wake-up words from appearing between smart devices, and ensure the user's accurate control of smart devices. To improve the user experience.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Human Computer Interaction (AREA)
  • Physics & Mathematics (AREA)
  • Acoustics & Sound (AREA)
  • Multimedia (AREA)
  • Quality & Reliability (AREA)
  • Signal Processing (AREA)
  • Artificial Intelligence (AREA)
  • Telephonic Communication Services (AREA)

Abstract

本文公开了一种设备控制方法、云端设备、智能设备、计算机介质及设备，应用于云端设备的方法包括：获取处于同一语音接收区域内的多个智能设备的唤醒词，其中，多个智能设备均能通过各自的唤醒词被唤醒；检测多个智能设备的唤醒词中是否有相同的唤醒词；在多个智能设备的唤醒词中有相同唤醒词的情况下，生成修改相同唤醒词的指令。本发明可有效识别处于同一语音接收区域内的使用相同唤醒词的多个智能设备，主动修改或通知智能设备修改唤醒词，防止智能设备间出现相同唤醒词的情况，保障用户对智能设备的准确控制，提高用户使用体验。

Description

设备控制方法、云端设备、智能设备、计算机介质及设备 技术领域
本发明实施例涉及互联网技术领域，尤其涉及设备控制方法、云端设备、智能设备、计算机介质及设备。
背景技术
随着技术的发展，各种类型的智能家居设备逐渐得到广泛的应用。现有的智能家居设备一般都支持语音唤醒功能，例如：使用者发出唤醒词，智能家居设备使用语音采集装置采集到语音信号后，识别此语音信号的内容为此智能家居设备设置的唤醒词后，便可以自动开启此智能家居设备。但随着家庭中智能家居设备的增大，可能多个智能家居设备会使用同一唤醒词，这样就容易造成使用者无法准确语音控制其想控制的设备，为使用者造成困扰，影响指令的有效执行。
发明内容
为了解决上述技术问题，本文提供了一种设备控制方法、云端设备、智能设备、计算机介质及设备。
本文提供的设备控制方法，应用于云端设备，包括：
获取处于同一语音接收区域内的多个智能设备的唤醒词，其中，多个智能设备均能通过各自的唤醒词被唤醒；
检测多个智能设备的唤醒词中是否有相同的唤醒词；
在多个智能设备的唤醒词中有相同唤醒词的情况下，生成修改相同唤醒词的指令。
上述设备控制方法还具有以下特点：
生成修改相同唤醒词的指令之后，方法还包括：根据唤醒词更新规则确定相同唤醒词。
上述设备控制方法还具有以下特点：
唤醒词更新规则包括对唤醒词增加具有顺序关系的后缀词。
上述设备控制方法还具有以下特点：
获取处于同一语音接收区域内的多个智能设备的唤醒词的同时，还包括：
获取处于同一语音接收区域内的多个智能设备所属路由器的硬件地址；
检测多个智能设备的唤醒词中是否有相同的唤醒词，包括：
根据多个智能设备所属路由器的硬件地址检测多个智能设备中是否有属于相同路由器的智能设备；
在多个智能设备中有属于相同路由器的智能设备的情况下，检测属于相同路由器的智能设备的唤醒词中是否有相同的唤醒词；
在多个智能设备的唤醒词中有相同唤醒词的情况下，生成修改相同唤醒词的指令，包括：
在属于相同路由器的智能设备的唤醒词中有相同的唤醒词的情况下，生成修改相同唤醒词的指令。
本文提供的一种计算机可读存储介质，此存储介质上存储有计算机程序，程序被处理器执行时实现上述方法的步骤。
本文提供的一种计算机设备，其中，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，处理器执行程序时实现上述方法的步骤。
本文提供的云端设备，包括：
第一获取模块，用于获取处于同一语音接收区域内的多个智能设备的唤醒词，其中，多个智能设备均能通过各自的唤醒词被唤醒；
第一检测模块，用于检测多个智能设备的唤醒词中是否有相同的唤醒词；
第一生成模块，用于在多个智能设备的唤醒词中有相同唤醒词的情况下，生成修改相同唤醒词的指令。
上述云端设备还具有以下特点：
还包括第二获取模块，用于在第一获取模块获取处于同一语音接收区域内的多个智能设备的唤醒词时，获取处于同一语音接收区域内的多个智能设备所属路由器的硬件地址；
第一检测模块包括第二检测模块，用于根据多个智能设备所属路由器的硬件地址检测多个智能设备中是否有属于相同路由器的智能设备，在多个智能设 备中有属于相同路由器的智能设备的情况下，检测属于相同路由器的智能设备的唤醒词中是否有相同的唤醒词；
第一生成模块包括第二生成模块，用于在属于相同路由器的智能设备的唤醒词中有相同的唤醒词的情况下，生成修改相同唤醒词的指令。
本文提供的设备控制方法，应用于智能设备，包括：
向云端设备发送智能设备的唤醒词，其中，智能设备能通过唤醒词被唤醒；
在云端设备检测到与智能设备处于同一语音接收区域内的其他智能设备的唤醒词与智能设备的唤醒词相同时，接收并存储云端设备发送的新唤醒词或接收云端设备发送的修改智能设备的唤醒词的指令。
上述设备控制方法还具有以下特点：
接收云端设备发送的修改智能设备的唤醒词的指令之后，方法还包括：根据唤醒词更新规则确定新唤醒词或接收并存储用户根据指令设置的新唤醒词。
上述设备控制方法还具有以下特点：
在新唤醒词确定之后，此方法还包括：根据新唤醒词更改预构建的语音识别解码网络。
上述设备控制方法还具有以下特点：
根据新唤醒词更改预构建的语音识别解码网络，包括：
确定新唤醒词对应的三音素节点序列，根据预构建的声学模型中的三音素与模型输出状态索引值的对应关系确定三音素节点序列中各个三音素对应的状态索引值，构成状态索引值序列；其中，预构建的声学模型为采用第一准则和第二准则训练全音素、垃圾音素和噪声音素得到的模型，第一准则是最大似然估计准则，第二准则是以最小音素错误准则为优化准则的区分性训练准则；
将预构建的语音识别解码网络中原唤醒词对应的状态索引值序列替换为新唤醒词对应的状态索引值序列。
上述设备控制方法还具有以下特点：
根据新唤醒词更改预构建的语音识别解码网络之后，此方法还包括：
接收用户输入的语音数据；从语音数据中提取得到语音特征；使用语音识别解码网络对语音特征进行解码，得到解码结果；在解码结果包括新唤醒词的 情况下，执行唤醒操作。
上述设备控制方法还具有以下特点：
向云端设备发送智能设备的唤醒词的同时，还包括：向云端设备发送智能设备所属路由器的硬件地址；
在云端设备检测到与智能设备处于同一语音接收区域内的其他智能设备的唤醒词与智能设备的唤醒词相同时，接收并存储云端设备发送的新唤醒词或接收云端设备发送的修改智能设备的唤醒词的指令，包括：
在云端设备检测到与智能设备处于同一语音接收区域内且路由器的硬件地址相同的其他智能设备的唤醒词与智能设备的唤醒词相同时，接收并存储云端设备发送的新唤醒词或接收云端设备发送的修改智能设备的唤醒词的指令。
本文提供的另一种计算机可读存储介质，存储介质上存储有计算机程序，程序被处理器执行时实现上述方法的步骤。
本文提供的另一种计算机设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，处理器执行程序时实现上述方法的步骤。
本文提供的智能设备，包括：
第一发送模块，用于向云端设备发送智能设备的唤醒词，其中，智能设备能通过唤醒词被唤醒；
第一接收模块，用于在云端设备检测到与智能设备处于同一语音接收区域内的其他智能设备的唤醒词与智能设备的唤醒词相同时，接收云端设备发送的新唤醒词或接收云端设备发送的修改智能设备的唤醒词的指令。
上述智能设备还具有以下特点：
智能设备还包括：
第二接收模块用于在第一接收模块接收云端设备发送的修改智能设备的唤醒词的指令后，接收并存储用户根据指令设置的新唤醒词；
唤醒词更新模块用于在第一接收模块接收云端设备发送的修改智能设备的唤醒词的指令后，根据唤醒词更新规则确定新唤醒词。
上述智能设备还具有以下特点：
智能设备还包括网络更新模块，用于在新唤醒词确定之后，根据新唤醒词 更改预构建的语音识别解码网络；
网络更新模块包括计算模块和网络维护模块；
计算模块，用于确定新唤醒词对应的三音素节点序列，根据预构建的声学模型中的三音素与模型输出状态索引值的对应关系确定三音素节点序列中各个三音素对应的状态索引值，构成状态索引值序列；
网络维护模块，用于将预构建的语音识别解码网络中原唤醒词对应的状态索引值序列替换为新唤醒词对应的状态索引值序列。
上述智能设备还具有以下特点：
智能设备还包括训练模块，用于采用第一准则和第二准则训练全音素、垃圾音素和噪声音素得到预构建的声学模型，第一准则是最大似然估计准则，第二准则是以最小音素错误准则为优化准则的区分性训练准则。
上述智能设备还具有以下特点：
智能设备还包括:
第二发送模块，用于在第一发送模块向云端设备发送智能设备的唤醒词的同时，向云端设备发送智能设备所属路由器的硬件地址；
第一接收模块包括第三接收模块，用于在云端设备检测到与智能设备处于同一语音接收区域内且路由器的硬件地址相同的其他智能设备的唤醒词与智能设备的唤醒词相同时，接收并存储云端设备发送的新唤醒词或接收云端设备发送的修改智能设备的唤醒词的指令。
本文可有效识别处于同一语音接收区域内的使用相同唤醒词的多个智能设备，生成修改相同唤醒词的指令，根据该指令后续可以主动修改或通知智能设备修改唤醒词，防止智能设备间出现相同唤醒词的情况，保障用户对智能设备的准确控制，提高用户使用体验。
附图说明
此处所说明的附图用来提供对本文的进一步理解，构成本申请的一部分，本文的示意性实施例及其说明用于解释本文，并不构成对本文的不当限定。在附图中：
图1是实施例一中应用于云端设备的设备控制方法的流程图；
图2是实施例一中云端设备的结构图；
图3是实施例二中应用于云端设备的设备控制方法的流程图；
图4是实施例二中智能设备的结构图。
具体实施方式
下文中将参考附图并结合实施例来详细说明本发明实施例。需要说明的是，在不冲突的情况下，本申请中的实施例及实施例中的特征可以相互组合。
本发明实施例中的智能设备典型的为智能家电设备。
实施例一
如图1所示，应用于云端设备的设备控制方法包括：
步骤101，获取处于同一语音接收区域内的多个智能设备的唤醒词，其中，多个智能设备均能通过各自的唤醒词被唤醒；
步骤102，检测多个智能设备的唤醒词中是否有相同的唤醒词；
步骤103，在多个智能设备的唤醒词中有相同唤醒词的情况下，生成修改相同唤醒词的指令。
本发明实施例可有效识别处于同一语音接收区域内的使用相同唤醒词的多个智能设备，生成修改相同唤醒词的指令，根据该指令后续可以主动修改或通知智能设备修改唤醒词，防止智能设备间出现相同唤醒词的情况，保障用户对智能设备的准确控制，提高用户使用体验。
本方法还可以包括步骤104：根据唤醒词更新规则确定相同唤醒词。其中，唤醒词更新规则包括对唤醒词增加具有顺序关系的后缀词。例如，对于原唤醒词为“开机”，更新后的唤醒词为“开机1”、“开机2”、“开机3”等；或者，更新后的唤醒词为“开机A”、“开机B”、“开机C”等。
根据上述描述可知，本实施例中更新唤醒词的方法主体可以是云端设备，本发明实施例提供不止一种的更新方式可以使用户根据使用需求选择相应的方式，提高用户的使用体验。
步骤102中获取处于同一语音接收区域内的多个智能设备的唤醒词的同时，还包括：获取处于同一语音接收区域内的多个智能设备所属路由器的硬件地址。
步骤102中，检测多个智能设备的唤醒词中是否有相同的唤醒词，包括：根据多个智能设备所属路由器的硬件地址检测多个智能设备中是否有属于相同路由器的智能设备；在多个智能设备中有属于相同路由器的智能设备的情况下，检测属于相同路由器的智能设备的唤醒词中是否有相同的唤醒词。
在多个智能设备的唤醒词中有相同唤醒词的情况下，步骤103中生成修改相同唤醒词的指令，包括：在属于相同路由器的智能设备的唤醒词中有相同的唤醒词的情况下，生成修改相同唤醒词的指令。
上述方法中，将处于同一局域网并且使用相同唤醒词的智能设备作为处于同一语音接收区域内智能设备，是考虑到一般处于同一局域网的智能设备一般是位于同一小型区域(例如家庭，公司等)内的设备，各智能设备之间的距离较小，容易因为使用相同唤醒词并且在使用户发出语音唤醒词时多个智能设备会接收到此唤醒词。
实施例一中还包括计算机可读存储介质，存储介质上存储有计算机程序，程序被处理器执行时实现上述方法的步骤。
实施例一中还包括一种计算机设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，处理器执行程序时实现上述方法的步骤。
如图2所示，实施例一中的云端设备包括：
第一获取模块，用于获取处于同一语音接收区域内的多个智能设备的唤醒词，其中，多个智能设备均能通过各自的唤醒词被唤醒；
第一检测模块，用于检测多个智能设备的唤醒词中是否有相同的唤醒词；
第一生成模块，用于在多个智能设备的唤醒词中有相同唤醒词的情况下，生成修改相同唤醒词的指令。
此云端设备还包括第二获取模块，用于在第一获取模块获取处于同一语音接收区域内的多个智能设备的唤醒词时，获取处于同一语音接收区域内的多个智能设备所属路由器的硬件地址。
第一检测模块包括第二检测模块，用于根据多个智能设备所属路由器的硬件地址检测多个智能设备中是否有属于相同路由器的智能设备，在多个智能设备中有属于相同路由器的智能设备的情况下，检测属于相同路由器的智能设备的唤醒词中是否有相同的唤醒词。
第一生成模块包括第二生成模块，用于在属于相同路由器的智能设备的唤醒词中有相同的唤醒词的情况下，生成修改相同唤醒词的指令。
此云端设备还包括：唤醒词修改模块，用于根据唤醒词更新规则确定相同唤醒词。唤醒词更新规则包括对唤醒词增加具有顺序关系的后缀词。例如，对于原唤醒词为“开机”，更新后的唤醒词为“开机1”、“开机2”、“开机3”等；或者，更新后的唤醒词为“开机A”、“开机B”、“开机C”等。
实施例二
如图3所示，应用于智能设备的设备控制方法包括：
步骤301，向云端设备发送智能设备的唤醒词，其中，智能设备能通过唤醒词被唤醒；
步骤302，在云端设备检测到与智能设备处于同一语音接收区域内的其他智能设备的唤醒词与智能设备的唤醒词相同时，接收并存储云端设备发送的新唤醒词或接收云端设备发送的修改智能设备的唤醒词的指令。
本方法中，步骤302中接收云端设备发送的修改智能设备的唤醒词的指令之后，此方法还包括：根据唤醒词更新规则确定新唤醒词或接收并存储用户根据指令设置的新唤醒词。
实施例二中智能设备接收到修改智能设备的唤醒词的指令后，更新唤醒词的方法主体可以是智能设备，也可以是用户，根据实施例一和实施例二，本发明实施例提供不止一种的更新方式可以使用户根据使用需求选择相应的方式，提高用户的使用体验。
为使云端设备能检测到多个智能设备的唤醒词中是否有相同的唤醒词，此智能设备向云端设备发送智能设备的唤醒词的同时，还包括：向云端设备发送智能设备所属路由器的硬件地址；在云端设备检测到与智能设备处于同一语音接收区域内的其他智能设备的唤醒词与智能设备的唤醒词相同时，步骤302中接收并存储云端设备发送的新唤醒词或接收云端设备发送的修改智能设备的唤醒词的指令具体包括：在云端设备检测到与智能设备处于同一语音接收区域内且路由器的硬件地址相同的其他智能设备的唤醒词与智能设备的唤醒词相同时，接收并存储云端设备发送的新唤醒词或接收云端设备发送的修改智能设备的唤醒词的指令。
步骤302中在新唤醒词确定之后，此方法还包括：根据新唤醒词更改预构建的语音识别解码网络。
此处需要说明的是，新唤醒词确定的方式可以是接收并存储云端设备发送的新唤醒词，也可以是智能设备根据唤醒词更新规则确定的新唤醒词或智能设备接收并存储用户根据指令设置的新唤醒词。
智能设备为实现唤醒词唤醒功能，预先构建声学模型，此预构建的声学模型为采用第一准则和第二准则训练全音素、垃圾音素和噪声音素得到的模型。其中，第一准则是最大似然估计(MaximumLikelihood Estimation，MLE)准则，第二准则是以最小音素错误(Minimum Phone Error，MPE)准则为优化准则的区分性训练准则。
构建声学模型时，根据发音变异规律使用全音素语音库中的全音素聚类成M个辅音音素和N个元音音素组成的垃圾音素库来训练垃圾模型；因为垃圾因素模型是由多个正常音素混合训练得到的，所以在解码时遇到唤醒词时输入语音的声学特征在垃圾音素上的得分是竞争不过在唤醒词对应的音素上的得分的，而遇到唤醒词之外的语音时因为垃圾音素中含相应的音素自然会竞争过唤醒词对应的音素。
此声学模型相比现有技术中常用的声学模型，增加了噪声音素的训练部分，可以提高声学模型的抗噪性能。并且，构建全音素声学模型很大的优点是在训练的时候不需要指定特定的唤醒词，只需要通过最大似然方法用数据集就可以产生一个覆盖全的三音素隐马尔可夫模型(Hidden Markov Model，HMM)。
在未获知唤醒词语音范围时，使用全音素语音库训练全音素声学模型，在预先获知唤醒词语音范围时，可以采集唤醒词语音范围的语音数据建立唤醒词语音库，并用此语音库来更新上述声学模型，使得上述声学模型对唤醒词有更好地匹配性。
声学模型训练成功后，保存有各个发音的三音素与模型输出状态索引值的对应关系。
智能设备还构建的解码网络，解码网络中包括并列的垃圾音素索引值节点、噪声音素索引值节点及当前使用的唤醒词对应的状态索引值序列。解码网络是一个可循环的网络，此循环的特点体现在可以从退出节点跳回到进入节点，这 样就可以覆盖持续的多个语音片断。
根据新唤醒词更改预构建的语音识别解码网络的方法包括：确定新唤醒词对应的三音素节点序列，根据预构建的声学模型中的三音素与模型输出状态索引值的对应关系确定三音素节点序列中各个三音素对应的状态索引值，构成状态索引值序列；将预构建的语音识别解码网络中原唤醒词对应的状态索引值序列替换为新唤醒词对应的状态索引值序列。
确定唤醒词对应的三音素节点序列的方式为使用每个字的音素通过结合前后音素组成三音素。在获取三音素节点序列时还可以使第一个三音素的第一个音素和最后一个三音素的最后一个音素包括静音音素(sil)。
如唤醒词是“你好电视“，则它对应的三音素节点序列为：
sil-n+i,n-i+h,i-h+ao,h-ao+d,ao-d+ian,d-ian+sh，ian-sh+i,sh-i+sil
本方法中，根据新唤醒词更改预构建的语音识别解码网络之后，使用此新唤醒词的方法包括：接收用户输入的语音数据；从语音数据中提取得到语音特征，例如此语音特征为美尔频率倒谱系数(Mel-Frequency Cepstrum Coefficient，MFCC)使用语音识别解码网络对语音特征进行解码，得到解码结果；在解码结果包括新唤醒词的情况下，执行唤醒操作。
实施例二中还提供了一种计算机可读存储介质，存储介质上存储有计算机程序，程序被处理器执行时实现上述方法的步骤。
实施例二中还提供了一种计算机设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，处理器执行程序时实现上述方法的步骤。
如图4所示，实施例二中的智能设备包括：
第一发送模块，用于向云端设备发送智能设备的唤醒词，其中，智能设备能通过唤醒词被唤醒；
第一接收模块，用于在云端设备检测到与智能设备处于同一语音接收区域内的其他智能设备的唤醒词与智能设备的唤醒词相同时，接收并存储云端设备发送的新唤醒词或接收云端设备发送的修改智能设备的唤醒词的指令。
此智能设备还包括第二接收模块或唤醒词更新模块。第二接收模块用于在第一接收模块接收云端设备发送的修改智能设备的唤醒词的指令后，接收并存 储用户根据指令设置的新唤醒词。唤醒词更新模块用于在第一接收模块接收云端设备发送的修改智能设备的唤醒词的指令后，根据唤醒词更新规则确定新唤醒词。
智能设备还包括网络更新模块，用于在新唤醒词确定之后，根据新唤醒词更改预构建的语音识别解码网络。
网络更新模块包括计算模块和网络维护模块。
计算模块用于确定新唤醒词对应的三音素节点序列，根据预构建的声学模型中的三音素与模型输出状态索引值的对应关系确定三音素节点序列中各个三音素对应的状态索引值，构成状态索引值序列。
网络维护模块用于将预构建的语音识别解码网络中原唤醒词对应的状态索引值序列替换为新唤醒词对应的状态索引值序列。
智能设备还包括训练模块，用于采用第一准则和第二准则训练全音素、垃圾音素和噪声音素得到上述预构建的声学模型，第一准则是最大似然估计准则，第二准则是以最小音素错误准则为优化准则的区分性训练准则。
智能设备还包括第二发送模块，用于在第一发送模块向云端设备发送智能设备的唤醒词的同时，向云端设备发送智能设备所属路由器的硬件地址。并且，第一接收模块包括第三接收模块用于在云端设备检测到与智能设备处于同一语音接收区域内且路由器的硬件地址相同的其他智能设备的唤醒词与智能设备的唤醒词相同时，接收并存储云端设备发送的新唤醒词或接收云端设备发送的修改智能设备的唤醒词的指令。
智能设备还包括第四接收模块、提取模块、解码模块和执行模块，其中第四接收模块，用于在网络更新模块根据新唤醒词更改预构建的语音识别解码网络之后，接收用户输入的语音数据；提取模块，用于从语音数据中提取得到语音特征；解码模块，用于使用语音识别解码网络对语音特征进行解码，得到解码结果；执行模块，用于在解码结果包括新唤醒词的情况下，执行唤醒操作。
对于本领域的技术人员来说，本发明实施例可以有各种更改和变化。凡在本发明实施例的精神和原则之内，所作的任何修改、等同替换、改进等，均应包含在本文的保护范围之内。
上面描述的内容可以单独地或者以各种方式组合起来实施，而这些变型方 式都在本发明的保护范围之内。
需要说明的是，本发明的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的本发明的实施例能够以除了在这里图示或描述的那些以外的顺序实施。
本领域普通技术人员可以理解，上文中所公开方法中的全部或某些步骤、系统、装置中的功能模块/单元可以被实施为软件、固件、硬件及其适当的组合。在硬件实施方式中，在以上描述中提及的功能模块/单元之间的划分不一定对应于物理组件的划分；例如，一个物理组件可以具有多个功能，或者一个功能或步骤可以由若干物理组件合作执行。某些组件或所有组件可以被实施为由处理器，如数字信号处理器或微处理器执行的软件，或者被实施为硬件，或者被实施为集成电路，如专用集成电路。这样的软件可以分布在计算机可读介质上，计算机可读介质可以包括计算机存储介质(或非暂时性介质)和通信介质(或暂时性介质)。如本领域普通技术人员公知的，术语计算机存储介质包括在用于存储信息(诸如计算机可读指令、数据结构、程序模块或其他数据)的任何方法或技术中实施的易失性和非易失性、可移除和不可移除介质。计算机存储介质包括但不限于RAM、ROM、EEPROM、闪存或其他存储器技术、CD-ROM、数字多功能盘(DVD)或其他光盘存储、磁盒、磁带、磁盘存储或其他磁存储装置、或者可以用于存储期望的信息并且可以被计算机访问的任何其他的介质。此外，本领域普通技术人员公知的是，通信介质通常包含计算机可读指令、数据结构、程序模块或者诸如载波或其他传输机制之类的调制数据信号中的其他数据，并且可包括任何信息递送介质。
工业实用性
本文可有效识别处于同一语音接收区域内的使用相同唤醒词的多个智能设备，主动修改或通知智能设备修改唤醒词，防止智能设备间出现相同唤醒词的情况，保障用户对智能设备的准确控制，提高用户使用体验。

Claims (21)

  1. 一种设备控制方法，所述方法应用于云端设备，包括：
    获取处于同一语音接收区域内的多个智能设备的唤醒词，其中，所述多个智能设备均能通过各自的唤醒词被唤醒；
    检测所述多个智能设备的唤醒词中是否有相同的唤醒词；
    在所述多个智能设备的唤醒词中有相同唤醒词的情况下，生成修改所述相同唤醒词的指令。
  2. 如权利要求1所述的设备控制方法，其中，生成修改所述相同唤醒词的指令之后，所述方法还包括：根据唤醒词更新规则确定所述相同唤醒词。
  3. 如权利要求2所述的设备控制方法，其中，
    所述唤醒词更新规则包括对唤醒词增加具有顺序关系的后缀词。
  4. 如权利要求1至3中任意一项所述的设备控制方法，其中，获取处于同一语音接收区域内的多个智能设备的唤醒词的同时，还包括：
    获取处于同一语音接收区域内的多个智能设备所属路由器的硬件地址；
    检测所述多个智能设备的唤醒词中是否有相同的唤醒词，包括：
    根据所述多个智能设备所属路由器的硬件地址检测所述多个智能设备中是否有属于相同路由器的智能设备；
    在所述多个智能设备中有属于相同路由器的智能设备的情况下，检测所述属于相同路由器的智能设备的唤醒词中是否有相同的唤醒词；
    在所述多个智能设备的唤醒词中有相同唤醒词的情况下，生成修改所述相同唤醒词的指令，包括：
    在所述属于相同路由器的智能设备的唤醒词中有相同的唤醒词的情况下，生成修改所述相同唤醒词的指令。
  5. 一种计算机可读存储介质，所述存储介质上存储有计算机程序，所述程序被处理器执行时实现权利要求1至4中任意一项所述方法的步骤。
  6. 一种计算机设备，其中，包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述处理器执行所述程序时实现权利 要求1至4中任意一项所述方法的步骤。
  7. 一种云端设备，包括：
    第一获取模块，用于获取处于同一语音接收区域内的多个智能设备的唤醒词，其中，所述多个智能设备均能通过各自的唤醒词被唤醒；
    第一检测模块，用于检测所述多个智能设备的唤醒词中是否有相同的唤醒词；
    第一生成模块，用于在所述多个智能设备的唤醒词中有相同唤醒词的情况下，生成修改所述相同唤醒词的指令。
  8. 如权利要求7所述云端设备，其中，
    还包括第二获取模块，用于在第一获取模块获取处于同一语音接收区域内的多个智能设备的唤醒词时，获取处于同一语音接收区域内的多个智能设备所属路由器的硬件地址；
    所述第一检测模块包括第二检测模块，用于根据所述多个智能设备所属路由器的硬件地址检测所述多个智能设备中是否有属于相同路由器的智能设备，在所述多个智能设备中有属于相同路由器的智能设备的情况下，检测所述属于相同路由器的智能设备的唤醒词中是否有相同的唤醒词；
    所述第一生成模块包括第二生成模块，用于在所述属于相同路由器的智能设备的唤醒词中有相同的唤醒词的情况下，生成修改所述相同唤醒词的指令。
  9. 一种设备控制方法，所述方法应用于智能设备，包括：
    向云端设备发送所述智能设备的唤醒词，其中，所述智能设备能通过所述唤醒词被唤醒；
    在所述云端设备检测到与所述智能设备处于同一语音接收区域内的其他智能设备的唤醒词与所述智能设备的唤醒词相同时，接收并存储所述云端设备发送的新唤醒词或接收所述云端设备发送的修改所述智能设备的唤醒词的指令。
  10. 如权利要求9所述的设备控制方法，其中，接收所述云端设备发送的修改所述智能设备的唤醒词的指令之后，所述方法还包括：
    根据唤醒词更新规则确定新唤醒词或接收并存储用户根据所述指令设置的新唤醒词。
  11. 如权利要求9或10所述的设备控制方法，其中，在新唤醒词确定之后，所述方法还包括：
    根据所述新唤醒词更改预构建的语音识别解码网络。
  12. 如权利要求11所述的设备控制方法，其中，根据所述新唤醒词更改预构建的语音识别解码网络，包括：
    确定所述新唤醒词对应的三音素节点序列，根据预构建的声学模型中的三音素与模型输出状态索引值的对应关系确定所述三音素节点序列中各个三音素对应的状态索引值，构成状态索引值序列；其中，所述预构建的声学模型为采用第一准则和第二准则训练全音素、垃圾音素和噪声音素得到的模型，所述第一准则是最大似然估计准则，所述第二准则是以最小音素错误准则为优化准则的区分性训练准则；
    将所述预构建的语音识别解码网络中原唤醒词对应的状态索引值序列替换为所述新唤醒词对应的状态索引值序列。
  13. 如权利要求11或12所述的设备控制方法，其中，根据所述新唤醒词更改预构建的语音识别解码网络之后，所述方法还包括：
    接收用户输入的语音数据；
    从所述语音数据中提取得到语音特征；
    使用所述语音识别解码网络对所述语音特征进行解码，得到解码结果；
    在所述解码结果包括所述新唤醒词的情况下，执行唤醒操作。
  14. 如权利要求9至12中任意一项所述的设备控制方法，其中，向云端设备发送所述智能设备的唤醒词的同时，还包括：
    向所述云端设备发送所述智能设备所属路由器的硬件地址；
    在所述云端设备检测到与所述智能设备处于同一语音接收区域内的其他智能设备的唤醒词与所述智能设备的唤醒词相同时，接收并存储所述云端设备发送的新唤醒词或接收所述云端设备发送的修改所述智能设备的唤醒词的指令，包括：
    在所述云端设备检测到与所述智能设备处于同一语音接收区域内且所述路由器的硬件地址相同的其他智能设备的唤醒词与所述智能设备的唤醒词相同 时，接收并存储所述云端设备发送的新唤醒词或接收所述云端设备发送的修改所述智能设备的唤醒词的指令。
  15. 一种计算机可读存储介质，所述存储介质上存储有计算机程序，所述程序被处理器执行时实现权利要求9至14中任意一项所述方法的步骤。
  16. 一种计算机设备，其中，包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述处理器执行所述程序时实现权利要求9至14中任意一项所述方法的步骤。
  17. 一种智能设备，包括：
    第一发送模块，用于向云端设备发送所述智能设备的唤醒词，其中，所述智能设备能通过所述唤醒词被唤醒；
    第一接收模块，用于在所述云端设备检测到与所述智能设备处于同一语音接收区域内的其他智能设备的唤醒词与所述智能设备的唤醒词相同时，接收并存储所述云端设备发送的新唤醒词或接收所述云端设备发送的修改所述智能设备的唤醒词的指令。
  18. 如权利要求17所述的智能设备，其中，所述智能设备还包括第二接收模块或唤醒词更新模块：
    所述第二接收模块用于在所述第一接收模块接收所述云端设备发送的修改所述智能设备的唤醒词的指令后，接收并存储用户根据所述指令设置的新唤醒词；
    所述唤醒词更新模块用于在所述第一接收模块接收所述云端设备发送的修改所述智能设备的唤醒词的指令后，根据唤醒词更新规则确定新唤醒词。
  19. 如权利要求17或18所述的智能设备，其中，所述智能设备还包括网络更新模块，用于在新唤醒词确定之后，根据所述新唤醒词更改预构建的语音识别解码网络；
    所述网络更新模块包括计算模块和网络维护模块；
    所述计算模块，用于确定所述新唤醒词对应的三音素节点序列，根据预构建的声学模型中的三音素与模型输出状态索引值的对应关系确定所述三音素节点序列中各个三音素对应的状态索引值，构成状态索引值序列；
    所述网络维护模块，用于将所述预构建的语音识别解码网络中原唤醒词对 应的状态索引值序列替换为所述新唤醒词对应的状态索引值序列。
  20. 如权利要求19所述的智能设备，其中，所述智能设备还包括训练模块，用于采用第一准则和第二准则训练全音素、垃圾音素和噪声音素得到所述预构建的声学模型，所述第一准则是最大似然估计准则，所述第二准则是以最小音素错误准则为优化准则的区分性训练准则。
  21. 如权利要求17所述的智能设备，其中，所述智能设备还包括：
    第二发送模块，用于在所述第一发送模块向云端设备发送所述智能设备的唤醒词的同时，向所述云端设备发送所述智能设备所属路由器的硬件地址；
    所述第一接收模块包括第三接收模块，用于在所述云端设备检测到与所述智能设备处于同一语音接收区域内且所述路由器的硬件地址相同的其他智能设备的唤醒词与所述智能设备的唤醒词相同时，接收并存储所述云端设备发送的新唤醒词或接收所述云端设备发送的修改所述智能设备的唤醒词的指令。
PCT/CN2017/116319 2017-12-15 2017-12-15 设备控制方法、云端设备、智能设备、计算机介质及设备 Ceased WO2019113911A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
PCT/CN2017/116319 WO2019113911A1 (zh) 2017-12-15 2017-12-15 设备控制方法、云端设备、智能设备、计算机介质及设备
CN201780097241.4A CN111386566A (zh) 2017-12-15 2017-12-15 设备控制方法、云端设备、智能设备、计算机介质及设备

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2017/116319 WO2019113911A1 (zh) 2017-12-15 2017-12-15 设备控制方法、云端设备、智能设备、计算机介质及设备

Publications (1)

Publication Number Publication Date
WO2019113911A1 true WO2019113911A1 (zh) 2019-06-20

Family

ID=66818893

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/116319 Ceased WO2019113911A1 (zh) 2017-12-15 2017-12-15 设备控制方法、云端设备、智能设备、计算机介质及设备

Country Status (2)

Country Link
CN (1) CN111386566A (zh)
WO (1) WO2019113911A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112929724A (zh) * 2020-12-31 2021-06-08 海信视像科技股份有限公司 显示设备、机顶盒及远场拾音唤醒控制方法
CN114220442A (zh) * 2022-01-27 2022-03-22 美的集团(上海)有限公司 智能家居系统的控制方法及智能家居系统
CN114697151A (zh) * 2022-03-15 2022-07-01 杭州控客信息技术有限公司 具有非语音唤醒功能的智能家居系统及其非语音唤醒方法

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116013280A (zh) * 2021-10-21 2023-04-25 海信集团控股股份有限公司 一种终端唤醒方法及装置
CN114155854B (zh) * 2021-12-13 2023-09-26 海信视像科技股份有限公司 语音数据的处理方法及装置
CN117354623A (zh) * 2023-12-04 2024-01-05 深圳市冠旭电子股份有限公司 拍照的控制方法、装置、电子设备及存储介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102999161A (zh) * 2012-11-13 2013-03-27 安徽科大讯飞信息科技股份有限公司 一种语音唤醒模块的实现方法及应用
CN103971678A (zh) * 2013-01-29 2014-08-06 腾讯科技（深圳）有限公司 关键词检测方法和装置
CN106469040A (zh) * 2015-08-19 2017-03-01 华为终端（东莞）有限公司 通信方法、服务器及设备
CN106960667A (zh) * 2017-03-08 2017-07-18 杭州联络互动信息科技股份有限公司 位置提醒方法、装置和系统
CN107134279A (zh) * 2017-06-30 2017-09-05 百度在线网络技术（北京）有限公司 一种语音唤醒方法、装置、终端和存储介质
CN107358954A (zh) * 2017-08-29 2017-11-17 成都启英泰伦科技有限公司 一种实时更换唤醒词的设备及方法

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
AU2011206677B9 (en) * 2010-01-12 2014-12-11 Fraunhofer-Gesellschaft Zur Foerderung Der Angewandten Forschung E.V. Audio encoder, audio decoder, method for encoding and decoding an audio information, and computer program obtaining a context sub-region value on the basis of a norm of previously decoded spectral values
CN103021409B (zh) * 2012-11-13 2016-02-24 安徽科大讯飞信息科技股份有限公司 一种语音启动拍照系统
US9996316B2 (en) * 2015-09-28 2018-06-12 Amazon Technologies, Inc. Mediation of wakeword response for multiple devices

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102999161A (zh) * 2012-11-13 2013-03-27 安徽科大讯飞信息科技股份有限公司 一种语音唤醒模块的实现方法及应用
CN103971678A (zh) * 2013-01-29 2014-08-06 腾讯科技（深圳）有限公司 关键词检测方法和装置
CN106469040A (zh) * 2015-08-19 2017-03-01 华为终端（东莞）有限公司 通信方法、服务器及设备
CN106960667A (zh) * 2017-03-08 2017-07-18 杭州联络互动信息科技股份有限公司 位置提醒方法、装置和系统
CN107134279A (zh) * 2017-06-30 2017-09-05 百度在线网络技术（北京）有限公司 一种语音唤醒方法、装置、终端和存储介质
CN107358954A (zh) * 2017-08-29 2017-11-17 成都启英泰伦科技有限公司 一种实时更换唤醒词的设备及方法

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112929724A (zh) * 2020-12-31 2021-06-08 海信视像科技股份有限公司 显示设备、机顶盒及远场拾音唤醒控制方法
CN114220442A (zh) * 2022-01-27 2022-03-22 美的集团(上海)有限公司 智能家居系统的控制方法及智能家居系统
CN114697151A (zh) * 2022-03-15 2022-07-01 杭州控客信息技术有限公司 具有非语音唤醒功能的智能家居系统及其非语音唤醒方法
CN114697151B (zh) * 2022-03-15 2024-06-07 杭州控客信息技术有限公司 具有非语音唤醒功能的智能家居系统及语音设备唤醒方法

Also Published As

Publication number Publication date
CN111386566A (zh) 2020-07-07

Similar Documents

Publication Publication Date Title
US11720326B2 (en) Audio output control
US11676585B1 (en) Hybrid decoding using hardware and software for automatic speech recognition systems
WO2019113911A1 (zh) 设备控制方法、云端设备、智能设备、计算机介质及设备
US12579981B1 (en) Target-device resolution
US11138974B2 (en) Privacy mode based on speaker identifier
US9934777B1 (en) Customized speech processing language models
KR102201937B1 (ko) 후속 음성 쿼리 예측
KR102733920B1 (ko) 자연어 프로세싱을 위한 시스템 및 방법
JP6448765B2 (ja) 対話装置、方法及びプログラム
US10170116B1 (en) Maintaining context for voice processes
JP6812843B2 (ja) 音声認識用コンピュータプログラム、音声認識装置及び音声認識方法
US11645468B2 (en) User data processing
CN111462748B (zh) 语音识别处理方法、装置、电子设备及存储介质
US11605376B1 (en) Processing orchestration for systems including machine-learned components
CN107644638A (zh) 语音识别方法、装置、终端和计算机可读存储介质
CN109741735B (zh) 一种建模方法、声学模型的获取方法和装置
US11646035B1 (en) Dialog management system
US11544504B1 (en) Dialog management system
JP7096199B2 (ja) 情報処理装置、情報処理方法、およびプログラム
CN111128172A (zh) 一种语音识别方法、电子设备和存储介质
US20180301144A1 (en) Electronic device, method for adapting acoustic model thereof, and voice recognition system
US12499876B2 (en) Multi-device speech processing
US20240212687A1 (en) Supplemental content output
US20220399013A1 (en) Response method, terminal, and storage medium
US10929601B1 (en) Question answering for a multi-modal system

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17934765

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 30.09.2020)

122 Ep: pct application non-entry in european phase

Ref document number: 17934765

Country of ref document: EP

Kind code of ref document: A1