CN103578468B - Adjusting method and an electronic apparatus in a speech recognition confidence threshold - Google Patents

Adjusting method and an electronic apparatus in a speech recognition confidence threshold Download PDF

Info

Publication number
CN103578468B
CN103578468B CN201210272154.7A CN201210272154A CN103578468B CN 103578468 B CN103578468 B CN 103578468B CN 201210272154 A CN201210272154 A CN 201210272154A CN 103578468 B CN103578468 B CN 103578468B
Authority
CN
China
Prior art keywords
value
confidence threshold
parameters
confidence
parameter
Prior art date
Application number
CN201210272154.7A
Other languages
Chinese (zh)
Other versions
CN103578468A (en
Inventor
戴海生
王茜莺
汪浩
Original Assignee
联想(北京)有限公司
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 联想(北京)有限公司 filed Critical 联想(北京)有限公司
Priority to CN201210272154.7A priority Critical patent/CN103578468B/en
Publication of CN103578468A publication Critical patent/CN103578468A/en
Application granted granted Critical
Publication of CN103578468B publication Critical patent/CN103578468B/en

Links

Abstract

本发明提供一种语音识别中置信度阈值的调整方法及电子设备。 The present invention provides a method for adjusting an electronic apparatus and a speech recognition confidence threshold. 该方法应用于一支持语音识别的电子设备中,所述语音识别当前的置信度阈值为第一值,所述方法包括:检测N个参数,获得N个检测结果,其中,N为大于等于1的整数;至少基于所述N个检测结果中的一个检测结果调整所述置信度阈值,使得所述置信度阈值从所述第一值变为第二值,其中,所述第二值为与所述第一值相同或不同的值。 The method is applied to an electronic device supports voice recognition, the voice recognition confidence threshold current value is a first value, the method comprising: detecting parameters of N, N detection results obtained, wherein, N is greater than or equal to 1 integer; at least based on a detection result of the detection of the N results in adjusting the confidence threshold, such that the confidence threshold value from the first value to a second value, wherein said second value and the first value is the same or different values.

Description

一种语音识别中置信度阈值的调整方法及电子设备 Adjusting method and an electronic apparatus in a speech recognition confidence threshold

技术领域 FIELD

[0001] 本发明涉及计算机技术领域,尤其涉及一种语音识别中置信度阈值的调整方法及电子设备。 [0001] The present invention relates to computer technology, and in particular relates to a method for adjusting an electronic apparatus and speech recognition confidence threshold.

背景技术 Background technique

[0002] 随着电子设备技术的发展,各种各样的电子设备进入了用户的生活,随着语音识别技术的发展,用户通过语音控制电子设备或者与电子设备进行语音交互的场景越来越多,给人们的生活带来了极大的便利。 [0002] With the development of electronics technology, a variety of electronic devices into the user's life, with the development of speech recognition technology, users control electronic devices by voice or voice interaction with electronic devices scenes more and more, to people's lives has brought great convenience.

[0003] 在语音控制或者语音交互的情形下,语音识别是很重要的一步,在语音识别的过程中,需要对识别结果进行确认,即将识别结果的置信度得分与置信度阈值进行比较,决定是否接受识别结果,当确认识别结果的置信度得分大于置信度阈值时,就接受该识别结果, 否则就拒绝该识别结果。 [0003] In the case of voice control or voice interaction, speech recognition is an important step in the process of speech recognition, it is necessary for the recognition result for confirmation, i.e. recognition result confidence score is compared with a confidence threshold is determined whether to accept the recognition result, the recognition result is confirmed when the confidence score is greater than the confidence threshold, the recognition result is accepted, otherwise it rejects the recognition result.

[0004] 然而,本发明人在实现本发明实施例中的技术方案的过程中发现,现有技术中的置信度阈值不管在什么情况下都固定不变,例如在环境比较嘈杂时,由于语音信号受到噪声的污染,导致识别结果的置信度得分下降,所以如果还按照原本的高阈值来判定的话,就会增加漏报率,即错误拒绝的比例会增加;或者当前的置信度阈值设置的比较低,而环境比较安静,识别结果的置信度得分偏高,这时还按照低阈值来确认的话,就增加误报率,即错误接受的比例会增加,使得语音识别的性能下降。 [0004] However, in the process of the present invention implement the technical solution in the embodiments of the present invention found that the confidence threshold prior art In any case fixed, for example, when noisy environment, since the voice signal is subjected to noise pollution, resulting in recognition result confidence score decreased, so if it is in accordance with the original high threshold value to determine, it would increase the false negative rate, ie the proportion of false rejects will increase; or the current confidence threshold setting it is relatively low, and relatively quiet environment, confidence score of the recognition result is high, then according to further confirm the low threshold, then it increases the false alarm rate, i.e. the proportion of acceptable error will increase, so that the speech recognition performance degradation.

发明内容 SUMMARY

[0005] 本发明提供一种语音识别中置信度阈值的调整方法及电子设备,用以解决现有技术中存在的语音识别的置信度阈值固定不变,使得语音识别的性能较低的技术问题。 [0005] The present invention provides a method for adjusting an electronic apparatus and a speech recognition confidence threshold, the confidence threshold to solve the prior art speech recognition in the presence of fixed values, such lower performance speech recognition Technical Problem .

[0006] 本发明一方面提供了一种语音识别中置信度阈值的调整方法,应用于一支持语音识别的电子设备中,所述语音识别当前的置信度阈值为第一值,所述方法包括:检测N个参数,获得N个检测结果,其中,N为大于等于1的整数;至少基于所述N个检测结果中的一个检测结果调整所述置信度阈值,使得所述置信度阈值从所述第—值变为第二值,其中,所述第二值为与所述第一值相同或不同的值。 [0006] In one aspect the present invention provides a method for adjusting a speech recognition confidence threshold, a voice recognition applied to an electronic device, the current speech recognition confidence value of a first threshold value, the method comprising : detecting a parameter N, N detection results obtained, wherein, N is an integer of 1; at least based on a detection result of the detection of the N results in adjusting the confidence threshold, such that the confidence threshold from the said first - value to a second value, wherein the second value is different from the first value or the same value.

[0007] 优选地,所述检测N个参数,具体为:检测所述电子设备所处的环境噪声参数;检测所述电子设备所处的工作场景复杂参数;和/或检测语音识别后的待确认语句的长度参数。 [0007] Preferably, the N parameters detected, in particular: detecting ambient noise parameter of the electronic device is located; the electronic device which detects operating parameters of scene complexity; and / or detection of the speech recognition to be Ensure that the length parameter statement.

[0008] 优选地,所述至少基于所述N个检测结果中的一个检测结果调整所述置信度阈值, 具体包括:基于所述环境噪声参数、环境噪声参数和置信度阈值的对应关系进行调整;基于所述工作场景复杂参数、工作场景复杂参数和置信度阈值的对应关系进行调整;和/或基于所述长度参数、长度参数和置信度阈值的对应关系进行调整。 [0008] Preferably, at least based on the adjustment of the confidence threshold detecting a detection result of said N results, comprises: a correspondence relationship is adjusted based on the ambient noise parameters, noise parameters and environmental confidence threshold ; scene complexity based on the operating parameters, operating parameter and a corresponding relationship between a complex scene confidence threshold is adjusted; and / or adjusted based on the correspondence relationship between the length parameter and the length parameter of the confidence threshold.

[0009] 优选地,在所述检测N个参数之前,所述方法还包括:接收第一语音输入;识别所述第一语音输入,获得第一识别结果。 [0009] Preferably, prior to detecting the N parameters, said method further comprising: receiving a first speech input; identifying the first speech input to obtain a first recognition result.

[0010] 优选地,基于所述第二值判断是否接受所述第一识别结果。 [0010] Preferably, the second value based on a first determination whether to accept the result of the recognition.

[0011] 本发明一实施例还提供一种电子设备,支持语音识别,所述语音识别当前的置信度阈值为第一值,所述电子设备包括:电路板;检测芯片,电性连接于所述电路板,用于检测N个参数,获得N个检测结果,其中,N为大于等于1的整数;处理芯片,设置在所述电路板上, 用于至少基于所述N个检测结果中的一个检测结果调整所述置信度阈值,使得所述置信度阈值从所述第一值变为第二值,其中,所述第二值为与所述第一值相同或不同的值。 [0011] The present invention further provides an embodiment of an electronic device, voice recognition, confidence threshold current value of the first speech recognition, the electronic device comprising: a circuit board; detection chip, electrically connected to the said circuit board, for detecting a parameter N, N detection results obtained, wherein, N is an integer of 1; and a processing chip, disposed on said circuit board, at least based on the N detection results adjusting a result of the confidence threshold detector, such that the confidence threshold value from the first value to a second value, wherein the second value is the value of the same or different from the first value.

[0012] 优选地,所述检测芯片具体用于检测所述电子设备所处的环境噪声参数;检测所述电子设备所处的工作场景复杂参数;和/或检测语音识别后的待确认语句的长度参数。 [0012] Preferably, the detecting chip is configured to ambient noise parameter detecting said electronic device is located; scene complexity detecting the operating parameters of the electronic device is located; and to be post / or confirmation sentence detecting speech recognition length parameter. [0013] 优选地,所述处理芯片具体用于基于所述环境噪声参数、环境噪声参数和置信度阈值的对应关系调整所述置信度阈值;基于所述工作场景复杂参数、工作场景复杂参数和置信度阈值的对应关系调整所述置信度阈值;和/或基于所述长度参数、长度参数和置信度阈值的对应关系调整所述置信度阈值。 [0013] Preferably, the chip is configured to adjust the correspondence relationship between confidence threshold parameter based on the ambient noise, ambient noise parameters and the confidence threshold; scene complexity based on the operating parameters, operating parameters and scene complexity adjusting the corresponding relationship between the confidence threshold confidence threshold value; and / or length based on the parameter, and the length parameter confidence threshold adjusting the correspondence between the confidence threshold.

[0014] 优选地,所述电子设备还包括:一声音采集单元,用于在所述检测芯片检测N个参数之前,接收第一语音输入;语音识别芯片,用于识别所述第一语音输入,获得第一识别结果。 [0014] Preferably, the electronic device further comprising: a sound collecting means for detecting, prior to said detecting chip N parameters, receiving a first input speech; speech recognition chip for speech input identifying the first obtain a first recognition result.

[0015] 优选地,所述语音识别芯片具体还用于基于所述第二值判断是否接受所述第一识别结果。 [0015] Preferably, the voice recognition chip further specifically based on the second value for determining whether to accept the first recognition result.

[0016] 本发明实施例中提供的一个或多个技术方案,至少具有如下技术效果或优点: [0016] One or more of the technical solutions provided in the embodiments of the present invention, at least the following technical effects or advantages:

[0017] 本发明一实施例采用实时检测一个或多个参数(例如环境、场景、语句本身),获得一个或多个检测结果,然后根据这些结果中的一个或多个检测结果对置信度阈值进行调整,使得置信度阈值能够从第一值变为第二值,其中,第二值为与第一值相同或不同的值。 [0017] The embodiment of the present invention, an embodiment employs real-time detection of one or more parameters (e.g. the environment, the scene, the statement itself), one or more detection results obtained, then the result according to one or more of these detection results of a confidence threshold is adjusted such that the confidence threshold can be changed to a second value from a first value, wherein the second value is the same as the first value or a different value. 如此一来,置信度阈值能够根据不同的环境、场景或者不同的语句变为适应环境、场景或者语句的置信度阈值,所以使得语音识别率更高,语音识别的性能更好。 Thus, the confidence threshold can be changed to adapt to the environment, the scene or a confidence threshold based on the statement of the different environments, different scene or statements, so that a higher rate of speech recognition, speech recognition performance better.

[0018] 进一步,本发明一实施例中还基于调整过的置信度阈值判断是否接受识别结果, 即先实时调整置信度阈值,然后根据调整后的置信度阈值进行判断识别结果的是否可信, 所以对识别结果的判断更合理,更准确。 [0018] Further, according to the present invention, an embodiment is also based on the adjusted confidence threshold is determined whether to accept the recognition result, i.e., the first time adjustment of the confidence threshold, and then determines a recognition result based on the confidence threshold after adjustment is trusted embodiment, it is judged that the recognition result is more reasonable, and more accurate.

附图说明 BRIEF DESCRIPTION

[0019]图1为本发明一实施例中的控制电子设备的方法流程图; [0019] FIG. 1 is a flowchart in the embodiment of a method of controlling an electronic device of the present embodiment of the invention;

[0020]图2为本发明一实施例中的电子设备的功能框图。 [0020] Figure 2 a functional block diagram of an electronic device according to an embodiment of the present invention.

具体实施方式 Detailed ways

[0021] 本发明提供一种语音识别中置信度阈值的调整方法及电子设备,用以解决现有技术中存在的语音识别的置信度阈值固定不变,使得语音识别的性能较低的技术问题。 [0021] The present invention provides a method for adjusting an electronic apparatus and a speech recognition confidence threshold, the confidence threshold to solve the prior art speech recognition in the presence of fixed values, such lower performance speech recognition Technical Problem .

[0022] 本发明实施例中的技术方案为解决上述的技术问题,总体思路如下: [0022] The technical solutions in the embodiments of the present invention is to solve the above technical problem, the general idea is as follows:

[0023] 通过实时检测一个或多个参数(例如环境、场景、语句本身),获得一个或多个检测结果,然后根据这些结果中的一个或多个检测结果对置信度阈值进行调整,使得置信度阈值能够从第一值变为第二值,其中,第二值为与第一值相同或不同的值。 [0023] The real-time detection of one or more parameters (e.g. the environment, the scene, the statement itself), obtained one or more detection results, and then adjust the confidence threshold results according to one or more of these detection results, so that the confidence the second threshold value can be changed from a first value, wherein the second value is the same as the first value or a different value. 如此一来,置信度阈值能够根据不同的环境、场景或者不同的语句变为适应环境、场景或者语句的置信度阚值,所以使得语音识别率更高,语音识别的性能更好。 Thus, the confidence threshold can be changed to adapt to the environment, the scene or confidence value Kan statement depending on the environment, different scene or statements, so that a higher rate of speech recognition, speech recognition performance better.

[0024]为了更好的理解上述技术方案,下面将结合说明书附图以及具体的实施方式对上述技术方案进行详细的说明。 [0024] For a better understanding of the above technical solutions, in conjunction with the following drawings and specific embodiments of the above technical solutions described in detail.

[0025] 本发明一实施例提供一种语音识别中置信度阈值的调整方法,应用于一支持语音识别的电子设备上,该电子设备例如是手机、平板电脑、笔记本电脑等电子设备,该语音识别当前的置信度阈值为第一值,例如为80。 [0025] The embodiment of the present invention provides a method of adjusting a confidence of speech recognition threshold, the voice recognition is applied to an electronic device, the electronic device such as a mobile phone, a tablet PC, notebook computers and other electronic equipment, the speech recognize the current value of a first confidence threshold value, for example 80.

[0026] 请参考图1,该方法包括: [0026] Please refer to FIG. 1, the method comprising:

[0027] 步骤101:检测N个参数,获得N个检测结果,其中,N为大于等于1的整数; [0027] Step 101: detecting a parameter N, N detection results obtained, wherein, N is an integer of 1;

[0028] 步骤102:至少基于N个检测结果中的一个检测结果调整置信度阈值,使得置信度阈值从第一值变为第二值,其中,第二值为与第一值相同或不同的值。 [0028] Step 102: adjusting at least based on a detection result of the detection results of N confidence threshold, such a confidence threshold from a first value to a second value, wherein the second value is the same or different from the first value value.

[0029] 其中,在步骤101中,检测N个参数,在具体实施过程中,N个参数的类别,即需要检测哪些参数,可以是用户通过一用户界面事先设置好的,例如对置信度得分影响比较大的环境噪声参数、工作场景复杂参数、和/或待确认语句的长度参数。 [0029] wherein, in the step 101, the parameter N is detected, the specific implementation process, the class of N parameters, i.e. parameters which need to detection, the user may be set well in advance by a user interface, for example, confidence score Effect of relatively large ambient noise parameters, operating parameters of scene complexity and / or length of the parameters to be confirmed statement. 在其他实施例中,也可以是其他影响置信度得分的参数,本领域技术人员可以根据实际需要进行设置。 In other embodiments, it may be other parameters affect the confidence score, one skilled in the art can be provided according to actual needs.

[0030] 在本实施例中,检测N个参数,具体可以是检测电子设备所处的环境噪声参数;检测电子设备所处的工作场景复杂参数;和/或测语音识别后的待确认语句的长度参数。 [0030] In the present embodiment, the detection of the N parameters, specific parameters may detect ambient noise of the electronic device is located; work in scene complexity parameter detection electronic device is located; and to be post / or confirmation sentence prediction speech recognition length parameter. 即可以只检测其中一个参数,也可以检测上述所有参数。 I.e., only one parameter can be detected, all of the above parameters may be detected.

[0031] 其中,环境噪声参数可以是噪声的分贝,或者是当接收一语音输入时,检测该语音输入的信噪比,当信噪比或分贝比较大时,说明对该语音输入的识别结果的得分会比较低, 所以这时需要检测环境噪声参数。 [0031] wherein ambient noise parameter may be a noise decibels, or when receiving a speech input, the speech input signal to noise ratio is detected, or when the SNR is relatively large dB, indicating the recognition result of the speech input the score will be relatively low, so when the need to detect ambient noise parameters.

[0032] 而工作场景复杂参数可以是事先对电子设备的每个工作场景复杂度进行评估,针对每个工作场景,都对应一个复杂系数,该复杂系数即可作为复杂参数,电子设备可以判断自己当前处于哪个工作场景,进而获得该工作场景下的复杂参数。 [0032] The working parameters may be pre-complex scene for each scene complexity evaluate operation of the electronic device, for each scene work, corresponds to a complex coefficient, the complex coefficient as a complex parameters to the electronic device based on its own which scenes work is currently in, and then get the complex parameters of the work scene.

[0033] 而待确认语句的长度参数,也可以事先进行训练,即训练词表中的每个语句的语句长度值,可以根据语句所包含的音素的个数来确定,或者根据包含的汉字的个数来确定。 [0033] and the length parameter statement to be confirmed, it can be trained in advance, i.e., the length of the training sentence word value for each statement in the table may be determined according to the number of phonemes contained in the sentence, or according to the Chinese characters comprising The number is determined. [0034] 进一步,还训练N个参数和置信度阈值之间的对应关系,在本实施例中,继续以N个参数为上述三个参数为例进行说明。 [0034] Further, also the correspondence between the training parameters N and confidence threshold, in this embodiment, N parameters continue to be an example for the three parameters mentioned above will be described.

[0035]首先,当环境噪声参数为分贝或者信噪比的时候,对于噪声分贝而言,可以确定一个分贝范围出来,该分贝范围可以是该电子设备可能在的最安静的环境分贝到最嘈杂的环境的环境分贝,也可以是其他的分贝范围,然后在这个分贝范围内针对每个分贝作相同的或不同的语音识别训练,计算对应分贝下的语音识别的识别结果的置信度值,最后会获得在该分贝下的一系列置信度值,然后可以取这一系列置信度值的平均值,作为对应分贝的置信度阈值,最终会形成一个分贝与置信度值的对应关系表。 [0035] First, when the environment parameter is noise or signal to noise ratio in decibels, dB for the noise, it may determine one out decibel range, the decibel range of the electronic device may be in the quietest possible environment to db noisiest environmental dB environment, may be other decibel range, then the range within which decibel dB for the same or different for each speech recognition training, calculate the confidence value of the recognition result of the speech recognition corresponding to decibels, and finally will receive a series of confidence in the dB value, then averaging the series of confidence values, confidence threshold as corresponding to decibels, dB will eventually form a correspondence table between confidence values. 当然,在实际应用时,也可以取这一系列置信度值的最低值,也可以取这一系列置信度值的最高值,或者可以取中间的某个值置信度值,而使得高于或等于该置信度值的置信度值的比例达到80%或者其他比例, 例如,训练得到的一系列的置信度值为80、81、82、78、79,那这时取79,所以根据统计学分析来看,如果将置信度阈值设置为79,那么在该分贝下,将会有80%的识别结果都会被接受。 Of course, in practice, may take the lowest value of the series of confidence values ​​can also take the highest value of the series of confidence value, or may take a value of an intermediate confidence value, such that above or confidence value is equal to the ratio of the confidence values ​​of 80%, or other ratio, for example, to obtain a series of training confidence value 80,81,82,78,79, 79 that take time, so according to the statistical analysis, if the confidence threshold is set to 79, then at this dB, there will be 80% of the recognition result will be accepted. [0036] 进一步,可以对该分贝与置信度值对应关系表进行进一步分析,会获得一个分贝与置信度阈值之间的函数关系式,当然也可以直接对分贝和置信度值进行多次训练,获得分贝与置信度阈值之间的函数关系式。 [0036] Further, the decibel may correspond to a confidence value table for further analysis, will get a functional relationship between the confidence threshold decibels, dB, and of course also be a confidence value for multiple training directly, obtaining a functional relationship between dB and the confidence threshold.

[0037]类似的,当环境噪声参数为信噪比的时候,可以训练0至1的彳目卩栄比的枢围卜的置信度阈值,针对范围内每个信噪比,都会获得一系列的置信度值,然后可以取这一系列置信度值的平均值,作为对应信噪比的置信度阈值,最终会形成一个分贝与置信度阈值的对应关系表。 [0037] Similarly, when the ambient noise when the signal to noise ratio parameter, the left foot can be trained to 0-1 mesh Jie confidence threshold Koei pivot around BU ratio, signal to noise ratio for each of the range, will receive a series of the confidence value, then averaging the series of confidence values, as the corresponding confidence threshold SNR, decibels eventually form a correspondence table of the confidence threshold. 当然,也可以如分贝的情况,根据在该信噪比下置信度值的分布情况取其他的置信度阈值。 Of course, as the case may be decibels, taking other confidence threshold based on the distribution of the values ​​of the confidence in the SNR. »038] 进一步,同样也可以获得信噪比与置信度阈值之间的函数关系式。 »038] Further, the same function can be obtained and the relationship between the confidence threshold SNR.

[0039]另外,环境噪声变大,即分贝变大,信噪比下降,导致正常语音的置信度得分下降, 所以需要下调置信度阈值,避免错误拒绝增多,所以最后训练的结果会是置信度阈值随着分贝的增大、信噪比的下降而下降,进而可以减少错误拒绝,而具体下调的量可以通过上述的训练方法获得。 [0039] In addition, environmental noise increases, ie becomes larger decibel noise ratio decreased, resulting in a normal voice confidence scores drop, so need to be lowered confidence threshold to avoid false rejection increased, so the final result will be the training of confidence with the increase of the threshold dB decrease of the SNR decreases, and thus can reduce the false rejection, and the amount of the specific down-regulation can be obtained by the above training method.

[0040] 对于工作场景复杂参数和长度参数对应的置信度阈值也可以按照上述训练方式进行训练,例如语句的长度参数,对于具有同样长度参数的多个语句进行语音识别,并计算识别结果的置信度值,得到该长度参数下的置信度值分布情况,进而可以得到对应的置信度阈值或者是长度参数和置信度阈值的函数关系式。 [0040] parameter and a length parameter corresponding to the confidence threshold can be performed for the work scene complexity according to the above-described training methods training, e.g. length parameter statement for a plurality of statements having the same length parameter for speech recognition, and calculates the recognition result confidence values, the distribution of confidence values ​​obtained in the length parameter, and thus can be obtained as a function of the confidence threshold of formula or a corresponding length parameter and the confidence threshold.

[0041] 另外,工作场景的变化,例如进入一个复杂的场景,语音识别的置信度得分可能会变高也可能会变低,所以置信度阈值可能会调高也可能会调低,以及调整的量可以通过上述的训练方式获得,使得置信度阈值和工作场景匹配,即使得错误拒绝和错误接受减少。 [0041] Further, the scene change work, eg, into a complex scene, speech recognition confidence score can become higher may also become low, so the confidence threshold may be lowered may increase, and the adjustment the amount of training can be obtained by the above-described embodiment, such a confidence threshold and a scene matching work, have even reduced false rejection and false acceptance. 对于待确认语句的长度的变化而言,与工作场景的变化情形类似,在此不再赘述。 For varying lengths of statements to be confirmed, the scene changes similar to the case of the work, which is not repeated herein.

[0042] 上述介绍的各种训练方法,既可以是出厂前厂商训练好的模型,或者第三方厂商训练好的模型,也可以是电子设备根据具体使用情况,逐渐训练的过程,例如每次在语音识别时,电子设备可以检测N个参数,并记录N个检测结果,然后再对应记录在这些检测结果下,语音识别结果的置信度的得分,然后就会获得一系列置信度值,即获得一个置信度值分布,电子设备进行统计分析、或者计算,例如按照取这一系列置信度值的平均值的原则或者取中间的某个值置信度值,而使得高于或等于该置信度值的置信度值的比例达到80%或者其他比例的原则,或者其他原则,自动建立并更新参数与置信度阈值的对应关系表或者函数关系式。 Various training method [0042] described above, may be a model trained manufacturer before delivery, or third-party vendors trained models may also be an electronic device in accordance with the specific use, gradual training process, for example, each time speech recognition, the electronic device may detect the N parameters, and record the N detection results, and then the corresponding at these detection results, the confidence score of the recognition result of the speech recording, and then it will obtain a series of confidence values, i.e. is obtained a confidence value distribution, statistical analysis of the electronic device, or calculated, for example according to the principle of taking the average of the series of confidence value or confidence value takes a value intermediate, such that the confidence value is greater than or equal to principle confidence value ratio of 80% or other ratios, or other principles to automatically create and update a correspondence table or a function formula parameters and the confidence threshold. 如此一来,电子设备可以通过学习、训练,自动更新对应关系表或者函数关系式, 使得调整后的置信度阈值更加符合实际情况,从而也使得语音识别的性能得到提高。 Thus, the electronic device may be by learning, training, automatically updates the correspondence table or the function formula, so that the confidence threshold adjusted more in line with the actual situation, which also makes the speech recognition performance is improved.

[0043]上述分别介绍了不同的参数各自的训练方法,但是也可以将不同的参数综合起来一起训练,最后得到多个参数和置信度阈值的对应关系,包括对应表或者函数关系式。 [0043] The different parameters are introduced each training methods, but may be different parameters together with training, and finally obtain the corresponding relationship between the plurality of parameters and the confidence threshold, comprising a correspondence table or a function formula. 进一步,对于其他参数也可以按照上述训练方式进行训练,为了说明书的简洁,在此不再赘述。 Further, other parameters may also be trained in accordance with the above-described training methods, the description for brevity, are not repeated herein. [0044]当然,上述训练方式只为举例,并非用于限制本发明,在具体实施过程中,本领域技术还可以采用其他训练方式进行训练。 [0044] Of course, the above-described training methods of example only, not intended to limit the present invention, in the specific implementation, those skilled in the embodiment may also employ other training training.

[0045] 在步骤1〇2中,至少基于N个检测结果中的一个检测结果调整置信度阈值,在一实施例中,步骤S102具体可以包括: [0045] In step 1〇2, adjusting a confidence threshold based on a detection result of at least the N detection results, in one embodiment, step S102 may include:

[0046] 基于N个检测结果中的至少一个检测结果,查询N个参数中至少一个参数的参数和置信度阈值的对应关系表,其中,N个参数中至少一个参数与N个检测结果中的至少一个检测结果分别对应; [0046] Based on the at least one detector of the N detection results of the results of the query correspondence table parameters and the confidence threshold value N parameters of at least one parameter, wherein, the N parameters at least one parameter of the N detection results of respectively corresponding to the at least one detection result;

[0047]在参数和置信度阈值的对应关系表中确定出与至少一个检测结果对应的置信度阈值; [0047] determined that the detection result of at least one confidence threshold corresponding to the correspondence table and the parameters of the confidence threshold;

[0048] 将与至少一个检测结果对应的置信度阈值设置为语音识别的置信度阈值。 [0048] The confidence threshold setting corresponding to the detection result of at least one confidence threshold speech recognition.

[0049] 具体来说,就是直接根据检测结果查询前述实例中所描述的对应关系表,因为每个参数或多个参数对应一个置信度阈值,只要将该对应关系表中的置信度阈值设为语音识别的置信度阈值即可。 [0049] Specifically, the query is a direct correspondence table described in the foregoing examples according to the detection result, since each parameter or plurality of parameters corresponding to a confidence threshold, as long as the correspondence relation table confidence threshold is set to speech recognition confidence threshold value. 调整后的置信度阈值可能和调整前的相同,也可能不相同。 Confidence threshold may be adjusted before the adjustment and the same, it may not be the same.

[0050] 在另一实施例中,步骤S102具体可以包括: [0050] In another embodiment, step S102 may include:

[0051] 基于N个检测结果中的至少一个检测结果,获取N个参数中至少一个参数的参数和置信度阈值之间的函数关系式,其中,N个参数中至少一个参数与N个检测结果中的至少一个检测结果分别对应; [0051] Based on at least a detection result of the N detection results, obtaining a function of the relationship between the N parameters at least one parameter and confidence threshold parameter, wherein, the N parameters at least one parameter of the N detection results at least one of the detection results correspond respectively;

[0052]将至少一个检测结果代入所述函数关系式进行计算,获得一计算置信度阈值; [0053]将该计算置信度阈值设置为语音识别的置信度阈值。 [0052] The at least one detection result is substituted into the function formula is calculated to obtain a calculated confidence threshold; [0053] The calculated value is set as the confidence threshold confidence threshold speech recognition.

[0054]具体来说,是通过上述实例中所描述的函数关系式,计算出该检测结果对应的置信度阈值,然后将该置信度阈值设置为语音识别的置信度阈值即可。 [0054] More specifically, by the functional relationship described in the above example, the detection result is calculated corresponding to the confidence threshold, then the confidence threshold is set to a speech recognition confidence threshold value. 调整后的置信度阈值可能和调整前的相同,也可能不相同。 Confidence threshold may be adjusted before the adjustment and the same, it may not be the same.

[0055] 在具体实施过程中,继续沿用上面的例子,具体可以是:基于环境噪声参数、环境噪声参数和置信度阈值的对应关系进行调整;基于工作场景复杂参数、工作场景复杂参数和置信度阈值的对应关系进行调整;和/或基于长度参数、长度参数和置信度阈值的对应关系进行调整。 [0055] In a specific embodiment the process, continue with the example above, may specifically be: adjusted ambient noise parameters, ambient noise parameters correspondence and confidence threshold based; based on the operating scenario complex parameters, the working scene complexity parameter and confidence a correspondence relationship is adjusted threshold; and / or length adjustment parameters corresponding relationship, and the length parameter based on the confidence threshold. 即可以只根据其中一个参数进行调整,也可以根据所有参数进行调整。 I.e., may be adjusted based on only one parameter may be adjusted in accordance with all the parameters.

[0056] 其中,对于环境参数(参数为分贝的情况)和工作场景复杂参数而言,因为是可以预先知道的,所以可以预先根据环境参数和工作场景复杂参数来调整置信度阈值,即将置信度阈值调整为与电子设备所处的环境和工作场景相适应,使得语音识别的误报率和漏报率都比较低,即错误接受和错误拒绝的情况发生的比例都比较低。 [0056] wherein, for environmental parameters (situation parameters dB) and the working scene complexity parameters, as is known in advance, it is possible in advance to adjust the confidence threshold based on environmental parameters and working scene complexity parameter, i.e. the confidence adjusting the threshold compatible with the working environment and scenarios in which the electronic device, so that the speech recognition rate of false positive and false negative rates are relatively low, i.e. where the proportion of false acceptance and false rejection occurs relatively low.

[0057]而对于环境参数(参数为待确认语句的信噪比的情况)和待确认语句的长度参数而言,因为是要分析待确认语句之后或者是识别待确认语句之后,再根据信噪比参数或者长度参数来调整置信度阈值。 For the length parameter [0057] For environmental parameters (parameters for the case statement to be confirmed SNR) and the statements to be confirmed, because the analysis is to be confirmed after the statement or statements following the identification to be confirmed, then in accordance with the signal to noise than the parameter or parameters to adjust the length of the confidence threshold.

[0058]当然,在实际运用时,也可以是在接收一语音输入之后,对该语音输入进行识别之后,再根据所有参数对置信度阈值进行调整。 [0058] Of course, in practical application, it may be received after a speech input, after identifying the voice input, and then adjust the confidence threshold based on all the parameters.

[0059]进一步,可以根据调整后的置信度阈值,对识别结果进行确认,判断是否要接受该识别结果,因为判断的依据是根据各种参数调整后的置信度阈值,所以使得漏报率和误报率都降低了,所以提高了语音识别的性能。 [0059] Further, according to the confidence threshold after adjustment of the recognition result, and determines whether to accept the recognition result, because the basis for judgment is a confidence threshold the various parameter adjustment, so that the false negative rate, and false positives are reduced, so improving speech recognition performance.

[0060]具体来说,例如,环境参数是由分贝来表征的,而且当前分贝数是90,表示电子设备目前处于一个嘈杂的环境中,例如马路上,表示语音信息被噪声污染的比较严重,进而导致语音识别的置信度得分会比较低,例如,这时用户输入了一个语音信息“我是小明”,经过语音识别,得到的识别结果同样也是“我是小明”,但是为了进一步确认该识别结果是否可信,则计算该识别结果的置信度得分,计算之后得知识别结果的置信度得分为60分,如果是按照现有技术的方法,假如置信度阈值始终固定在80分,然后将识别结果的置信度得分60 分与置信度阈值80进行比较,发现识别结果的置信度得分小于系统设定的置信度阈值,所以判定该结果不可信,所以就不会进一步处理该语音信息,例如发送给别的电子终端,或者显示在显示单元上,但实际上这个识别 [0060] Specifically, for example, environmental parameters characterized in decibels, and the current is 90 decibels, showing the electronic device is currently in a noisy environment, for example on the road, the speech information is expressed serious noise pollution, leading to speech recognition confidence score will be relatively low, e.g., when the user inputs a voice message "I am Bob" through speech recognition, recognition result is also "I am Bob," but in order to further confirm the identification results are trusted, then calculate the confidence score of the recognition result, the recognition result that after the calculation of the confidence score of 60 points, if the method according to the prior art, if the confidence threshold is always fixed at 80 minutes, and then recognition result confidence score of 60 points compared with the confidence threshold 80, found confidence score of the recognition result is less than the confidence threshold set by the system, it is determined that the result credible, so that no further processing of the voice information, e.g. transmitted to the other electrical terminal or displayed on the display unit, but in fact the identification 果是可信的,却因为置信度阈值设置太高,而导致错误拒绝该识别结果。 If trusted, but because the confidence threshold is set too high, resulting in false rejection of the recognition result.

[0061]然而,通过本实施例中描述的置信度阈值调整方法,电子设备通过检测环境参数, 根据环境参数自动的将置信度阈值调整为与环境相适应的阈值,例如,检测到分贝数是90, 可以通过前述查表的方式或者将参数带入函数关系式中的方式,得到一个合理的置信度阈值,例如是59,然后将识别结果的置信度得分60分与置信度阈值59分进行比较,结果是识别结果的置信度得分大于置信度阈值,所以说明该识别结果是可信的,所以就会对该语音信息进行下一步处理。 [0061] However, the confidence threshold adjustment method of the present embodiment described, the electronic device by detecting environmental parameters, automatically the confidence threshold adjust the threshold value environmentally compatible according to environmental parameters, e.g., the detected number of decibels 90, by the table lookup or parameters into the functional relationship in a manner to give a reasonable confidence threshold, for example 59, then the recognition result confidence score of 60 points with the confidence threshold 59 minutes for comparison, the result is a recognition result confidence score is greater than the confidence threshold, thus indicating that the recognition result is reliable, it will process the voice information to the next step.

[0062]通过以上具体实例看出,通过本实施例中的置信度阈值的调整方法调整后的置信度阈值更合理,在语音识别中降低了错误拒绝的比例,即降低了漏报率。 [0062] Specific examples of the above seen, the confidence threshold is adjusted by adjusting the confidence threshold method of the present embodiment is more reasonable to reduce the proportion of false reject the speech recognition, i.e. reducing the false negative rate. 同样的道理,也可以降低错误接受的比例,降低误报率。 By the same token, it can also reduce the proportion of false acceptance and false positive rates.

[0063]因此,本实施例中的置信度阈值能够根据环境、工作场景、待确认语句的长度的变化而自适应调整,当然还包括其他对置信度得分影响比较大的其他参数,所以使得置信度阈值能够调整在一个合理的值上,减少错误接受和错误拒绝,使得语音识别的准确率更高, 语音识别性能更好。 [0063] Thus, the confidence threshold in the present embodiment can be depending on the environment, the working scene, of varying length to be confirmed statements adaptively adjusted, of course, other parameters other on confidence score greatest impact, so that the confidence threshold can be adjusted at a reasonable value, reducing false acceptance and false rejection, so that a higher rate of speech recognition accuracy, better speech recognition performance.

[0064]本发明一实施例中还提供一种电子设备,该电子设备例如是手机、平板电脑、笔记本电脑等电子设备,该电子设备支持语音识别,语音识别当前的置信度阈值为第一值。 [0064] In a further embodiment the present invention provides an electronic device, the electronic device such as a mobile phone, a tablet PC, notebook computers and other electronic devices, the electronic device supports voice recognition, voice recognition confidence threshold current value is a first value .

[0065]如图2所示,该电子设备包括:电路板201;检测芯片202,电性连接于电路板201,用于检测N个参数,获得N个检测结果,其中,N为大于等于1的整数;处理芯片203,设置在电路板201上,用于至少基于N个检测结果中的一个检测结果调整置信度阈值,使得置信度阈值从第一值变为第二值,其中,第二值为与第一值相同或不同的值。 [0065] As shown in FIG. 2, the electronic device comprising: a circuit board 201; detecting chip 202 electrically connected to the circuit board 201, for detecting a parameter N, N detection results obtained, wherein, N is greater than or equal to 1 integer; processing chip 203, provided on the circuit board 201, at least based on a detection result of the detection results of the N adjusted confidence threshold, such a confidence threshold from a first value to a second value, wherein the second different values ​​or value is identical to the first.

[0066] 其中,检测芯片202例如是分贝仪,用于检测电子设备输出的环境噪声,分贝,或者是包括:备份子芯片、傅里叶变换子芯片、滤波子芯片、傅里叶反变换子芯片、计算子芯片的芯片,具体检测过程为:通过麦克风录入第一语音信号,然后先通过备份子芯片将第一语音信号进行备份,生成一第一备份语音信号;然后第一语音信号通过傅里叶变换子芯片变化, 在频域通过滤波子芯片滤波,去除噪声,然后将消除噪声后的数据进行傅里叶反变换子芯片的处理,然后将第一备份语音信号和去除噪声的第一语音信号在计算子芯片进行计算, 得出环境参数信噪比。 [0066] wherein detecting a decibel meter, for example, chip 202, an electronic device for detecting ambient noise output, dB, or comprising: a backup sub-chip, chip sub Fourier transform, filtering sub-chip sub inverse Fourier transform , chip calculation sub-chips, the specific procedure involves: the first voice signal input through a microphone, and the first speech signal is first backed up by backup sub-chips, the first backup generating a speech signal; and a first speech signal by Fu Fourier transformation sub-chip variation, in the frequency domain by filtering the sub-chip filtering, noise removal, and then the noise elimination processing data inverse Fourier transform of the sub-chip, and then the first speech signal and the first backup removing noise speech signal is calculated in a computing sub-chip SNR obtained environmental parameters.

[0067] 在另一实施例中,检测芯片202还可以直接检测电子设备所处的工作场景。 [0067] In another embodiment, the detection chip 202 may also directly detect the electronic device in which the working scene.

[0068] 在另一实施例中,检测芯片202还可以是语音处理芯片,进而检测待确认语句的长度参数。 [0068] In another embodiment, the chip 202 may also be detected voice processing chip, to be detected and further confirmed the statement length parameter.

[0069] 在另一实施例中,检测芯片202包括以上描述的各种芯片,可以检测以上描述的各种参数。 [0069] In embodiments, the various chips 202 comprises a detection chip as described above In another embodiment, the various parameters described above may be detected.

[0070] 进一步,处理芯片203具体用于基于环境噪声参数、环境噪声参数和置信度阈值的对应关系调整置信度阈值;基于工作场景复杂参数、工作场景复杂参数和置信度阈值的对应关系调整置信度阈值;和/或基于长度参数、长度参数和置信度阈值的对应关系调整置信度阈值。 [0070] Further, the processing chip 203 is configured to adjust the confidence level threshold corresponds to the relationship between environmental noise parameters, ambient noise parameters and the confidence threshold based; based on the operating scene complexity correspondence relationship between adjustment parameters, the complexity parameter and confidence threshold operating scenarios Confidence threshold value; and / or adjust the confidence threshold corresponding relationship between the length of the parameter, and the length parameter based on the confidence threshold.

[0071] 进一步,处理芯片303可以是单独的处理芯片,也可以是集成在电子设备的中央处理器中。 [0071] Further, the processing chip 303 may be a separate processing chip, or may be integrated in the central processor of the electronic device.

[0072] 在一实施例中,电子设备还包括:一声音采集单元,用于在检测芯片202检测N个参数之前,接收第一语音输入;语音识别芯片,用于识别第一语音输入,获得第一识别结果。 [0072] In one embodiment, the electronic device further comprising: a sound collecting unit for detecting the chip 202 prior to the detection of the N parameters, receiving a first input speech; speech recognition chip for recognizing a first speech input to obtain a first recognition result. 声音采集单元,例如麦克风;语音识别芯片可以与处理芯片203是相同的芯片,也可以是不同的芯片。 Sound collection means, such as a microphone; speech recognition processing chip 203 and the chip may be the same chip or different chips.

[0073] 进一步,语音识别芯片具体还用于基于第二值判断是否接受第一识别结果。 [0073] Further, voice recognition chip further particular value based on a second determination whether to accept a first recognition result.

[0074]以上各实施例可以单独实施,也可以结合实施,技术人员可根据实际需要进行选择。 [0074] each of the above embodiments may be implemented individually embodiments, embodiments can also be combined, the skilled artisan can be selected according to actual needs.

[0075]前述图1实施例中的置信度阈值调整方法中的各种变化方式和具体实例同样适用于本实施例的电子设备,通过前述对置信度阈值调整方法的详细描述,本领域技术人员可以清楚的知道本实施例中电子设备的实施方法,所以为了说明书的简洁,在此不再详述。 [0075] the FIG. 1 embodiment variations embodiment and specific examples confidence threshold adjustment method embodiment in equally applicable to the present embodiment of the electronic device, by the foregoing detailed description of the method for adjusting the confidence threshold, those skilled in the art can know the method of an embodiment of an electronic device according to the present embodiment, the description for simplicity, not described in detail here. [0076] 本发明实施例中提供的一个或多个技术方案,至少具有如下技术效果或优点: [0077] 本发明一实施例采用实时检测一个或多个参数(例如环境、场景、语句本身),获得一个或多个检测结果,然后根据这些结果中的一个或多个检测结果对置信度阈值进行调整,使得置信度阈值能够从第一值变为第二值,其中,第二值为与第一值相同或不同的值。 [0076] One or more of the technical solutions provided in the embodiments of the present invention, having at least the following technical effects or advantages: [0077] embodiment of the present invention uses a real-time detection of one or more parameters (e.g. the environment, the scene, the statement itself) obtain one or more detection results, and then adjust the confidence threshold according to one or more of the results of these detection results, so that the confidence threshold can be changed to a second value from a first value, wherein the second value is with a first identical value or different values. 如此一来,置信度阈值能够根据不同的环境、场景或者不同的语句变为适应环境、场景或者语句的置信度阈值,所以使得语音识别率更高,语音识别的性能更好。 Thus, the confidence threshold can be changed to adapt to the environment, the scene or a confidence threshold based on the statement of the different environments, different scene or statements, so that a higher rate of speech recognition, speech recognition performance better.

[0078] 进一步,本发明一实施例中还基于调整过的置信度阈值判断是否接受识别结果, 即先实时调整置信度阚值,然后根据调整后的置信度阈值进行判断识别结果的是否可信, 所以对识别结果的判断更合理,更准确。 [0078] Further, an embodiment of the present invention is further based on the adjusted confidence threshold is determined whether to accept the recognition result, i.e., the first real-time adjustment confidence Kan value, and then determines a recognition result based on the confidence threshold the adjusted credibility , it is judged that the recognition result is more reasonable, and more accurate.

[0079]本领域内的技术人员应明白,本发明的实施例可提供为方法、系统、或计算机程序产品。 [0079] skilled in the art should understand that the embodiments of the present invention may provide a method, system, or computer program product. 因此,本发明可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。 Thus, embodiments of the present invention may be employed entirely hardware embodiment, an entirely software embodiment, or an embodiment in conjunction with the form of software and hardware aspects. 而且,本发明可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器和光学存储器等)上实施的计算机程序产品的形式。 Further, in the present invention may comprise one or more of which computer usable storage medium having computer-usable program code (including but not limited to disk storage, and optical storage) in the form of a computer program product implemented on.

[0080]本发明是参照根据本发明实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。 [0080] The present invention has been described in accordance with the method of Example of the present invention, apparatus (systems) and computer program products flowchart and / or block diagrams described. 应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。 It should be understood and implemented by computer program instructions and block, and the flowchart / or block diagrams each process and / or flowchart illustrations and / or block diagrams of processes and / or blocks. 可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器,使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。 These computer program instructions may be provided to a processor a general purpose computer, special purpose computer, embedded processor or other programmable data processing apparatus to produce a machine, such that the instructions executed by the processor of the computer or other programmable data processing apparatus generating in a device for implementing the flow chart or more flows and / or block diagram block or blocks in a specified functions. [0081]这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中,使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品,该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。 [0081] These computer program instructions may also be stored in a computer can direct a computer or other programmable data processing apparatus to function in a particular manner readable memory produce an article of manufacture such that the storage instruction means comprises a memory in the computer-readable instructions the instruction means implemented in a flowchart or more flows and / or block diagram block or blocks in a specified function.

[0082]这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上,使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理,从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。 [0082] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps on the computer or other programmable apparatus to produce a computer implemented so that the computer or other programmable apparatus execute instructions to provide processes for implementing a process or flows and / or block diagram block or blocks a function specified step.

[0083]显然,本领域的技术人员可以对本发明进行各种改动和变型而不脱离本发明的精神和范围。 [0083] Obviously, those skilled in the art can make various modifications and variations to the invention without departing from the spirit and scope of the invention. 这样,倘若本发明的这些修改和变型属于本发明权利要求及其等同技术的范围之内,则本发明也意图包含这些改动和变型在内。 Thus, if these modifications and variations of the present invention fall within the claims of the invention and the scope of equivalents thereof, the present invention intends to include these modifications and variations.

Claims (10)

1. 一种语音识别中置信度阈值的调整方法,应用于一支持语音识别的电子设备中,所述语音识别当前的置信度阈值为第一值,其特征在于,所述方法包括: 检测N个参数,获得N个检测结果,其中,N为大于等于1的整数; 至少基于所述N个检测结果中的一个检测结果调整所述置信度阈值,使得所述置信度阈值从所述第一值变为第二值,其中,所述第二值为与所述第一值相同或不同的值; 将语音识别结果的置信度得分与所述第二值进行比较,并根据比较结果确定是否接受所述语音识别结果。 1. A speech recognition method for adjusting the value of the confidence threshold, a voice recognition applied to an electronic device, the current value of the first speech recognition confidence threshold value, characterized in that the method comprises: detecting N parameters, to obtain the N detection results, where, N is an integer of 1; adjusted based on at least a detection result of the confidence threshold value of the N detection results, so that the confidence threshold value from the first value to a second value, wherein the second value is different from the first value or the same value; the speech recognition result confidence score is compared with the second value, and determines whether the result of the comparison receiving the speech recognition result.
2. 如权利要求1所述的方法,其特征在于,所述检测N个参数,具体为: 检测所述电子设备所处的环境噪声参数; 检测所述电子设备所处的工作场景复杂参数;和/或检测语音识别后的待确认语句的长度参数。 2. The method according to claim 1, wherein said detecting N parameters, specifically: detecting the ambient noise parameters of the electronic device is located; the electronic device which detects operating parameters of scene complexity; and to be post / or detection of speech recognition parameters confirm the length of the statement.
3. 如权利要求2所述的方法,其特征在于,所述至少基于所述N个检测结果中的一个检测结果调整所述置信度阈值,具体包括: 基于所述环境噪声参数、环境噪声参数和置信度阈值的对应关系进行调整; 基于所述工作场景复杂参数、工作场景复杂参数和置信度阈值的对应关系进行调整; 和/或基于所述长度参数、长度参数和置信度阈值的对应关系进行调整。 3. The method according to claim 2, wherein said at least one detection based on a detection result of said N results in adjusting the confidence threshold, comprises: a parameter ambient noise, ambient noise parameters based on and a correspondence relationship between the confidence threshold is adjusted; adjusted based on the operating scene complexity correspondence relationship parameter, complex parameters and the confidence threshold operating scenarios; and / or corresponding relation based on the length parameters, the length parameter and confidence threshold to adjust.
4. 如权利要求1所述的方法,其特征在于,在所述检测N个参数之前,所述方法还包括: 接收第一语音输入; 识别所述第一语音输入,获得第一识别结果。 4. The method according to claim 1, wherein, prior to said detecting the N parameter, the method further comprising: receiving a first speech input; identifying the first speech input to obtain a first recognition result.
5. 如权利要求4所述的方法,其特征在于,基于所述第二值判断是否接受所述第一识别结果。 5. The method according to claim 4, wherein the second value is determined based on whether to accept the first recognition result.
6. —种电子设备,支持语音识别,所述语音识别当前的置信度阈值为第一值,其特征在于,所述电子设备包括: 电路板; 检测芯片,电性连接于所述电路板,用于检测N个参数,获得N个检测结果,其中,N为大于等于1的整数; 处理芯片,设置在所述电路板上,用于至少基于所述N个检测结果中的一个检测结果调整所述置信度阈值,使得所述置信度阈值从所述第一值变为第二值,其中,所述第二值为与所述第一值相同或不同的值;将语音识别结果的置信度得分与所述第二值进行比较,并根据比较结果确定是否接受所述语音识别结果。 6. - electronic devices, voice recognition, the voice recognition of the current value of a first confidence threshold value, wherein the electronic device comprises: a circuit board; detection chip, electrically connected to the circuit board, for detecting a parameter N, N detection results obtained, wherein, N is an integer of 1; and a processing chip, disposed on said circuit board, for adjusting at least based on a detection result of the detection result of the N the confidence threshold, such that the confidence threshold value from the first value to a second value, wherein the second value is different from the first value or the same value; the confidence of the speech recognition result score is compared with the second value, and determines whether to accept the result of the speech recognition result of the comparison.
7. 如权利要求6所述的电子设备,其特征在于,所述检测芯片具体用于检测所述电子设备所处的环境噪声参数;检测所述电子设备所处的工作场景复杂参数;和/或检测语音识别后的待确认语句的长度参数。 7. The electronic device according to claim 6, wherein said detecting chip is configured to detect ambient noise parameter of the electronic device is located; the electronic device which detects operating parameters of scene complexity; and / speech recognition or detection to be confirmed after the length parameter statement.
8.如权利要求7所述的电子设备,其特征在于,所述处理芯片具体用于基于所述环境噪声参数、环境噪声参数和置信度阈值的对应关系调整所述置信度阈值;基于所述工作场景复杂参数、工作场景复杂参数和置信度阈值的对应关系调整所述置信度阈值;和/或基于所述长度参数、长度参数和置信度阈值的对应关系调整所述置信度阈值。 8. The electronic apparatus according to claim 7, wherein said chip is configured to adjust based on the correspondence relationship between the confidence threshold parameter ambient noise, ambient noise parameters and the confidence threshold; based on the complex scenes work parameters, the corresponding relationship between the confidence threshold adjustment parameter and the complex confidence threshold operating scenarios; and / or adjusted based on a corresponding relationship between the length of the confidence threshold parameter and the length parameter of the confidence threshold.
9.如权利要求6所述的电子设备,其特征在于,所述电子设备还包括: 一声音采集单元,用于在所述检测芯片检测N个参数之前,接收第一语音输入; 语音识别芯片,用于识别所述第一语音输入,获得第一识别结果。 9. The electronic apparatus according to claim 6, wherein the electronic device further comprising: a sound collecting means for detecting, prior to said detecting chip N parameters, receiving a first input speech; speech recognition chip for identifying the first speech input to obtain a first recognition result.
10.如权利要求9所述的电子设备,其特征在于,所述语音识别芯片具体还用于基于所述第二值判断是否接受所述第一识别结果。 10. The electronic apparatus according to claim 9, wherein the voice recognition chip further specifically based on the second value for determining whether to accept the first recognition result.
CN201210272154.7A 2012-08-01 2012-08-01 Adjusting method and an electronic apparatus in a speech recognition confidence threshold CN103578468B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201210272154.7A CN103578468B (en) 2012-08-01 2012-08-01 Adjusting method and an electronic apparatus in a speech recognition confidence threshold

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201210272154.7A CN103578468B (en) 2012-08-01 2012-08-01 Adjusting method and an electronic apparatus in a speech recognition confidence threshold

Publications (2)

Publication Number Publication Date
CN103578468A CN103578468A (en) 2014-02-12
CN103578468B true CN103578468B (en) 2017-06-27

Family

ID=50050121

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201210272154.7A CN103578468B (en) 2012-08-01 2012-08-01 Adjusting method and an electronic apparatus in a speech recognition confidence threshold

Country Status (1)

Country Link
CN (1) CN103578468B (en)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103065631B (en) * 2013-01-24 2015-07-29 华为终端有限公司 A speech recognition method, apparatus
CN103971680B (en) 2013-01-24 2018-06-05 华为终端(东莞)有限公司 The method of speech recognition species, means
CN106653010A (en) * 2015-11-03 2017-05-10 络达科技股份有限公司 Electronic apparatus and voice trigger method therefor
CN105702253A (en) * 2016-01-07 2016-06-22 北京云知声信息技术有限公司 Voice awakening method and device
CN106338924A (en) * 2016-09-23 2017-01-18 广州视源电子科技股份有限公司 Method and device for automatically adjusting operation parameter threshold of equipment
CN106384594A (en) * 2016-11-04 2017-02-08 湖南海翼电子商务股份有限公司 On-vehicle terminal for voice recognition and method thereof

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1264888A (en) * 1998-12-17 2000-08-30 索尼国际(欧洲)股份有限公司 Semi-monitoring speaker self-adaption
CN1434406A (en) * 2001-12-12 2003-08-06 微软公司 System and method for self-adaptive communication prioritization and route
CN1842842A (en) * 2003-08-29 2006-10-04 松下电器产业株式会社 Method and apparatus for improved speech recognition with supplementary information
CN101206857A (en) * 2006-12-19 2008-06-25 国际商业机器公司 Method and system for modifying speech processing arrangement
CN101272416A (en) * 2006-11-28 2008-09-24 通用汽车公司 Voice dialing using a rejection reference
CN101447185A (en) * 2008-12-08 2009-06-03 深圳市北科瑞声科技有限公司;深港产学研基地产业发展中心 Audio frequency rapid classification method based on content
US7657433B1 (en) * 2006-09-08 2010-02-02 Tellme Networks, Inc. Speech recognition accuracy with multi-confidence thresholds
CN101640042A (en) * 2008-07-29 2010-02-03 佳能株式会社 Information processing method and information processing apparatus
CN102136269A (en) * 2010-01-22 2011-07-27 微软公司 Speech recognition analysis via identification information
CN102194454A (en) * 2010-03-05 2011-09-21 富士通株式会社 Equipment and method for detecting key word in continuous speech

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8005668B2 (en) * 2004-09-22 2011-08-23 General Motors Llc Adaptive confidence thresholds in telematics system speech recognition
US8396715B2 (en) * 2005-06-28 2013-03-12 Microsoft Corporation Confidence threshold tuning

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1264888A (en) * 1998-12-17 2000-08-30 索尼国际(欧洲)股份有限公司 Semi-monitoring speaker self-adaption
CN1434406A (en) * 2001-12-12 2003-08-06 微软公司 System and method for self-adaptive communication prioritization and route
CN1842842A (en) * 2003-08-29 2006-10-04 松下电器产业株式会社 Method and apparatus for improved speech recognition with supplementary information
US7657433B1 (en) * 2006-09-08 2010-02-02 Tellme Networks, Inc. Speech recognition accuracy with multi-confidence thresholds
CN101272416A (en) * 2006-11-28 2008-09-24 通用汽车公司 Voice dialing using a rejection reference
CN101206857A (en) * 2006-12-19 2008-06-25 国际商业机器公司 Method and system for modifying speech processing arrangement
CN101640042A (en) * 2008-07-29 2010-02-03 佳能株式会社 Information processing method and information processing apparatus
CN101447185A (en) * 2008-12-08 2009-06-03 深圳市北科瑞声科技有限公司;深港产学研基地产业发展中心 Audio frequency rapid classification method based on content
CN102136269A (en) * 2010-01-22 2011-07-27 微软公司 Speech recognition analysis via identification information
CN102194454A (en) * 2010-03-05 2011-09-21 富士通株式会社 Equipment and method for detecting key word in continuous speech

Also Published As

Publication number Publication date
CN103578468A (en) 2014-02-12

Similar Documents

Publication Publication Date Title
CN104956688B (en) A speaker for equalizing the mobile device
KR101721303B1 (en) Voice activity detection in presence of background noise
EP2669889A2 (en) Method and apparatus for executing voice command in electronic device
CN104010147B (en) The method of automatically adjusting the volume of the audio playback system and audio playback devices
CN102318325B (en) Controlling an adaptation of a behavior of an audio device to a current acoustic environmental condition
CN103168455A (en) Mobile device having increased security that is less obtrusive
GB2346999A (en) Communication device for endpointing speech utterances
US20060085183A1 (en) System and method for increasing recognition accuracy and modifying the behavior of a device in response to the detection of different levels of speech
EP2774420A2 (en) Extraction and analysis of audio feature data
JP2016519784A (en) Apparatus and method for audio classification and processing
KR101768264B1 (en) A noise suppressing method and a noise suppressor for applying the noise suppressing method
JP2016507086A (en) Ambient noise root mean square (rms) detector
CN102469656A (en) Method, device and terminal for controlling indicator lamp
CN102752687A (en) Sound effect adjusting method for terminal device and terminal device
CN105210146B (en) Method and apparatus for controlling voice activation
KR101726208B1 (en) Volume leveler controller and controlling method
US20060100866A1 (en) Influencing automatic speech recognition signal-to-noise levels
WO2011141772A1 (en) Method and apparatus for processing an audio signal based on an estimated loudness
CN100504840C (en) Method for fast dynamic estimation of background noise
CN104978166A (en) Volume adjusting method and device
CN103761463B (en) An information processing method and an electronic device
KR20160009344A (en) Method and apparatus for recognizing whispered voice
CN102710838B (en) Volume regulation method and device as well as electronic equipment
US9921803B2 (en) Audio user interface apparatus and method
CN104731304B (en) Method and apparatus for adjusting the FM mode terminal cpu

Legal Events

Date Code Title Description
C06 Publication
GR01