CN103810997A - Method and device for determining confidence of voice recognition result - Google Patents
Method and device for determining confidence of voice recognition result Download PDFInfo
- Publication number
- CN103810997A CN103810997A CN201210459131.7A CN201210459131A CN103810997A CN 103810997 A CN103810997 A CN 103810997A CN 201210459131 A CN201210459131 A CN 201210459131A CN 103810997 A CN103810997 A CN 103810997A
- Authority
- CN
- China
- Prior art keywords
- arc
- confidence
- degree
- word
- competitive relation
- 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.)
- Granted
Links
Images
Abstract
The invention provides a method and a device for determining the confidence of a voice recognition result, wherein the method comprises the following steps: determining the confidence of each arc in a word graph obtained by decoding, and determining an optimal path in the word graph; determining an arc assembly T competitive to an arc Ai in the word graph for each arc Ai on the optimal path; determining an arc Aj from the arc assembly T competitive to an arc Ai when the confidence of a word expressed by the arc Ai, wherein the arc Aj and the arc Ai represent the same words, or the arc Aj and the arc assembly connected with the arc Aj are formed into a word expressed as same as the arc Ai; determining the confidence of the word represented by the arc Ai according to the confidence of the arc Ai and the confidence of the arc Aj or further according to the confidence of the arc connected with the arc Aj. When the method and the device are used for determining the confidence of the voice recognition result, the component factors of a compound word are considered, so that the confidence can further reflect the real state accurately.
Description
[technical field]
The present invention relates to the field of speech recognition in Computer Applied Technology, particularly a kind of method and apparatus of definite voice identification result degree of confidence.
[background technology]
In speech recognition, degree of confidence is used for representing that recognition result is the possibility of correct result, the possibility that the larger expression recognition result of value is correct result is higher, be the important evidence of carrying out speech recognition, definite method of voice identification result degree of confidence has directly affected the accuracy of speech recognition.
The degree of confidence of voice identification result determines it is mainly by the word figure (Aattice) of decoding generation is processed and obtained.Word figure is a kind of voice identification result form of expression of commonly using in recent years, and it represents multiple candidate result of decoding on a directed acyclic graph, in retaining many candidate informations, has saved storage space.In word figure, arc represents word, represents the annexation of word with node, and each word belongs to one from starting the path of node to end node.Wherein every arc in word figure can represent { W, A by a five-tuple
w, L
w, S
w, E
w, wherein W represents the word that arc is corresponding, A
wrepresent to produce the acoustics score of word W, L
wrepresent to produce the language score of word W, S
wrepresent to produce the start time of word W, E
wrepresent to produce the end time of word W.Fig. 1 is the example of a word figure, and in figure, <s> and </s> represent respectively path first symbol and path end mark.
The degree of confidence of existing voice recognition result is when definite, determine the degree of confidence of word according to optimal path, as shown in fig. 1, for " Chinese people ", because optimal path is arc " Chinese people ", therefore the degree of confidence of this word is arc " Chinese people's " degree of confidence.But in Chinese, a word can be made up of two other word, it is so-called compound word, corresponding to such word, formed by word " China " and " people " as " Chinese people ", the degree of confidence of existing voice recognition result determines that mode just ignored the constituent element of compound word, make recognition result degree of confidence can not reflect real situation, because the degree of confidence of recognition result may exert an influence in the self-adaptation adjustment process of follow-up acoustic model and language model, therefore also can bring impact to the accuracy of recognition result.
[summary of the invention]
In view of this, the invention provides a kind of method and apparatus of definite voice identification result degree of confidence, so that improve the accuracy of voice identification result degree of confidence.
Concrete technical scheme is as follows:
A method for definite voice identification result degree of confidence, the method comprises:
S1, determine the degree of confidence of every arc in the word figure that obtains of decoding, and optimal path in definite word figure;
S2, to every arc A on described optimal path
i, in word figure, determine and this arc A
ithere is the arc set T of competitive relation;
S3, at definite described arc A
irepresent the degree of confidence of word time, from described A
iexist in the arc set T of competitive relation and determine arc A
j, wherein arc A
jwith arc A
irepresent identical word, or arc A
jbe connected arc constitutes and arc A with it
irepresent identical word; In conjunction with arc A
iwith arc A
jdegree of confidence, or further combined with described arc A
jconnect arc degree of confidence determine arc A
ithe degree of confidence of the word representing.
According to one preferred embodiment of the present invention, in described step S1, the degree of confidence of every arc equals the value obtaining divided by the score sum in all paths in word figure through the score sum in all paths of this arc.
According to one preferred embodiment of the present invention, in described step S2, determine when whether two arcs exist competitive relation, in the following ways:
If two arc on the duration, exist overlapping, determine two arcs there is competitive relation; Or,
If two arc exists overlappingly on the duration, and the word that two arcs represent meets preset requirement in enunciative similarity, determines that two arcs exist competitive relation.
According to one preferred embodiment of the present invention, described S3 specifically comprises:
S31, initialization arc A
ithe degree of confidence of the word representing is arc A
idegree of confidence;
S32, from described arc A
iexist in the arc set T of competitive relation and select an arc not being selected;
S33, judge the arc selected whether with arc A
irepresent identical word, if so, by arc A
ithe degree of confidence currency that the degree of confidence of word representing is updated to this word adds the degree of confidence of the arc of selection, execution step S35; Otherwise, execution step S34;
S34, judge that whether the arc that is connected with it of arc of selecting combines and arc A
irepresent identical word, if so, in conjunction with arc A
iin the degree of confidence currency of the word representing and the combination of described arc, the degree of confidence of each arc is upgraded arc A
ithe degree of confidence of the word representing, execution step S35; Otherwise directly perform step S35;
S35, judge in described arc set T whether also have non-selected arc, if so, go to described step S32; Otherwise, finish arc A
ithe degree of confidence of the word representing is determined flow process.
According to one preferred embodiment of the present invention, described in step S34 in conjunction with arc A
iin the degree of confidence currency of the word representing and the combination of described arc, the degree of confidence of each arc is upgraded arc A
ithe degree of confidence of the word representing is specially:
By arc A
ithe degree of confidence currency that the degree of confidence of word representing is updated to this word adds the degree of confidence minimum value of each arc in the above arc combination.
A device for definite voice identification result degree of confidence, this device comprises:
Initial determining unit, for determining the degree of confidence of every arc of word figure that decoding obtains, and optimal path in definite word figure;
Set determining unit, for every arc A on described optimal path
i, in word figure, determine and this arc A
ithere is the arc set T of competitive relation;
Degree of confidence determining unit, at definite described arc A
irepresent the degree of confidence of word time, from described A
iexist in the arc set T of competitive relation and determine arc A
j, wherein arc A
jwith arc A
irepresent identical word, or arc A
jbe connected arc constitutes and arc A with it
irepresent identical word; In conjunction with arc A
iwith arc A
jdegree of confidence, or further combined with described arc A
jconnect arc degree of confidence determine arc A
ithe degree of confidence of the word representing.
According to one preferred embodiment of the present invention, described initial determining unit determines that the degree of confidence of every arc equals the value obtaining divided by the score sum in all paths in word figure through the score sum in all paths of this arc.
According to one preferred embodiment of the present invention, described set determining unit is in the time that whether definite two arcs there is competitive relation, in the following ways:
If two arc on the duration, exist overlapping, determine two arcs there is competitive relation; Or,
If two arc exists overlappingly on the duration, and the word that two arcs represent meets preset requirement in enunciative similarity, determines that two arcs exist competitive relation.
According to one preferred embodiment of the present invention, described degree of confidence determining unit specifically comprises:
Initialization subelement, for initialization arc A
ithe degree of confidence of the word representing is arc A
idegree of confidence, trigger arc chooser unit;
Arc chooser unit, for after being triggered from described arc A
iexist in the arc set T of competitive relation and select an arc not being selected;
First upgrades subelement, for judge the arc selected described arc chooser unit whether with arc A
irepresent identical word, if so, by arc A
ithe degree of confidence currency that the degree of confidence of word representing is updated to this word adds the degree of confidence of the arc of selection, triggers judgment sub-unit; Otherwise trigger second and upgrade subelement;
Second upgrades subelement, for judging that whether the arc that is connected with it of arc of selecting described arc chooser unit combines and arc A
irepresent identical word, if so, in conjunction with arc A
iin the degree of confidence currency of the word representing and the combination of described arc, the degree of confidence of each arc is upgraded arc A
ithe degree of confidence of the word representing, triggers judgment sub-unit; Otherwise directly trigger judgment sub-unit;
Judgment sub-unit, for judging whether described arc set T also exists non-selected arc, if so, triggers described arc chooser unit, otherwise finishes arc A
ithe degree of confidence of the word representing is determined flow process.
According to one preferred embodiment of the present invention, described second upgrades subelement renewal arc A
irepresent the degree of confidence of word time, specifically by arc A
ithe degree of confidence currency that the degree of confidence of word representing is updated to this word adds the degree of confidence minimum value of each arc in the above arc combination.
As can be seen from the above technical solutions, the present invention is in the time determining the degree of confidence of voice identification result, consider the constituent element of compound word, constitute the situation of a word for multiple words, the degree of confidence of this combined situation is also included in to the degree of confidence of word and determined, make degree of confidence reflect more exactly real conditions.
[accompanying drawing explanation]
Fig. 1 is an instance graph of word figure;
The method flow diagram that Fig. 2 provides for the embodiment of the present invention one;
The specific implementation process flow diagram of step 204 in Fig. 2 that Fig. 3 provides for the embodiment of the present invention one;
The structure drawing of device of definite voice identification result degree of confidence that Fig. 4 provides for the embodiment of the present invention two;
The structural drawing of the degree of confidence determining unit that Fig. 5 provides for the embodiment of the present invention two.
[embodiment]
In order to make the object, technical solutions and advantages of the present invention clearer, describe the present invention below in conjunction with the drawings and specific embodiments.
Embodiment mono-,
The method flow diagram that Fig. 2 provides for the embodiment of the present invention one, as shown in Figure 2, the method can specifically comprise the following steps:
Step 201: the degree of confidence of determining every arc in the word figure obtaining that decodes.
In this step, the degree of confidence of every arc equals the value obtaining divided by the score sum in all paths in word figure through the score sum in all paths of this arc, and wherein path must be divided into the acoustics score in this path and the summation of language score.
Still, take the figure of word shown in Fig. 1 as example, that supposes path " People's University " must be divided into 5, and path " China "-" people " must be divided into 3, and path " China "-" people " must be divided into 2, can obtain so:
The degree of confidence of arc " People's University " is
The degree of confidence of arc " China " is
The degree of confidence of arc " people " is
The degree of confidence of arc " people " is
Step 202: determine the optimal path in word figure.
Optimal path in the predicate figure of institute is exactly the highest path of score in all paths.
Above-mentioned steps 201 and step 202 are that prior art does not repeat them here, and in addition, above-mentioned steps 201 and step 202 also can be carried out simultaneously, also can successively carry out according to random order, and said sequence is only a kind of embodiment wherein.
Step 203: to every arc A on optimal path
i, in word figure, determine and this arc A
ithere is the arc set T of competitive relation.
In the time judging whether two arcs exist competitive relation, can determine according to time factor, if two arcs exist overlappingly on the duration, determine that two arcs exist competitive relation.For example, two arc A
1and A
2: A
1={ W
1, A
w1, L
w1, S
w1, E
w1, A
2={ W
2, A
w2, L
w2, S
w2, E
w2, if meet S
w2≤ (S
w1+ E
w1)/2 < E
w2, think arc A
2with arc A
1there is competitive relation.
In order to describe more accurately competitive relation, except according to time factor, also need the word that two arcs represent to meet just definite competitive relation that exists of preset requirement in enunciative similarity, wherein enunciative similarity can adopt the editing distance of syllable to embody, and also can adopt the Euclidean distance of acoustic model or language model to embody.
Step 204: determining every arc A on optimal path
irepresent the degree of confidence of word time, from arc A
iexist in the arc set T of competitive relation and determine arc A
j, wherein arc A
jwith arc A
irepresent identical word, or arc A
jbe connected arc constitutes and arc A with it
iidentical word, in conjunction with arc A
iand A
jdegree of confidence, or further combined with above-mentioned A
jconnect arc degree of confidence determine arc A
ithe degree of confidence of the word representing.
Particularly, this step can be for every arc A on optimal path
ithereby the concrete flow process of carrying out as shown in Figure 3 obtains the degree of confidence of the word that every arc represents respectively, comprises the following steps as shown in Figure 3:
Step 301: initialization arc A
ithe degree of confidence of the word representing is arc A
idegree of confidence.
Step 302: from this arc A
iexist in the arc set of competitive relation and select an arc not being selected.
Step 303: judge the arc selected whether with arc A
irepresent identical word, if so, execution step 304; Otherwise, execution step 305.
Step 304: by arc A
ithe degree of confidence currency that the degree of confidence of word representing is set to this word adds the degree of confidence of the arc of selection, execution step 307.
Step 305: judge that whether the arc that is connected with it of arc of selecting combines and arc A
irepresent identical word, if so, execution step 306; Otherwise, execution step 307.
Here the arc that the arc of selection can be expanded forward or expand to be backward connected with it combines.
Step 306: by arc A
ithe degree of confidence currency that the degree of confidence of word representing is updated to this word adds the minimum value of the degree of confidence of each arc in above-mentioned arc combination, execution step 307.
For example, in the figure of word shown in Fig. 1, arc " China " is an arc that has competitive relation with arc " Chinese people ", the arc combination of expanding backward connected arc " people " due to arc " China " also represents " Chinese people ", the degree of confidence of arc " Chinese people " is added after the minimum value between arc " China " and arc " people " to the degree of confidence of " Chinese people " using the value obtaining as word.
Adopt herein the degree of confidence currency of word to add that the degree of confidence minimum value of each arc in arc combination is a kind of preferred implementation that the present embodiment adopts, except this embodiment, can also adopt the mode such as degree of confidence mean value that adds each arc in arc combination such as the degree of confidence currency of word, just accuracy is not as adding that the mode of minimum value is high.
Step 307: judgement and this arc A
ihave in the arc set of competitive relation whether also have non-selected arc, if so, go to step 302; Otherwise finish arc A
ithe degree of confidence of the word representing is determined flow process.
So just can obtain on optimal path the degree of confidence that every arc represents word, this degree of confidence has comprised each word in compound word and has been decoded as respectively the situation of independent word, makes the degree of confidence to have reflected more exactly the possibility of this word as optimal identification result.
Be more than the detailed description that method provided by the present invention is carried out, below in conjunction with embodiment bis-, device provided by the invention be described in detail.
Embodiment bis-,
The structure drawing of device of definite voice identification result degree of confidence that Fig. 4 provides for the embodiment of the present invention two, as shown in Figure 4, this device can comprise: initial determining unit 400, set determining unit 410 and degree of confidence determining unit 420.
First initial determining unit 400 is determined the degree of confidence of every arc in the word figure that decoding obtains, and optimal path in definite word figure.Particularly, the degree of confidence of every arc equals the value obtaining divided by the score sum in all paths in word figure through the score sum in all paths of this arc, and wherein path must be divided into the acoustics score in this path and the summation of language score.Optimal path in word figure is exactly the highest path of score in all paths.
Then gather determining unit 410 to every arc A on optimal path
i, in word figure, determine and this arc A
ithere is the arc set T of competitive relation.
Wherein, set determining unit 410, can be in the following ways in the time that whether definite two arcs exist competitive relation: overlapping if two arcs exist on the duration, definite two arcs exist competitive relation; Or, overlapping if two arcs exist on the duration, and the word that two arcs represent meets preset requirement in enunciative similarity, determines that two arcs exist competitive relation.Wherein enunciative similarity can adopt the editing distance of syllable to embody, and also can adopt the Euclidean distance of acoustic model or language model to embody.
Last degree of confidence determining unit 420 is at definite arc A
irepresent the degree of confidence of word time, from A
iexist in the arc set T of competitive relation and determine arc A
j, wherein arc A
jwith arc A
irepresent identical word, or arc A
jbe connected arc constitutes and arc A with it
irepresent identical word; In conjunction with arc A
iwith arc A
j, or further combined with arc A
jconnect arc degree of confidence determine arc A
ithe degree of confidence of the word representing.
Below the concrete structure of degree of confidence determining unit 420 is described in detail, as shown in Figure 5, this degree of confidence determining unit 420 can specifically comprise: initialization subelement 421, arc chooser unit 422, the first renewal subelement 423, the second renewal subelement 424 and judgment sub-unit 425.
Wherein initialization subelement 421, for initialization arc A
ithe degree of confidence of the word representing is arc A
idegree of confidence, then trigger arc chooser unit 422.
First upgrades subelement 423, for judge the arc selected arc chooser unit 422 whether with arc A
irepresent identical word, if so, by arc A
ithe degree of confidence currency that the degree of confidence of word representing is updated to this word adds the degree of confidence of the arc of selection, triggers judgment sub-unit 425; Otherwise trigger second and upgrade subelement 424.
Second upgrades subelement 424, for judging that whether arc that arc that arc chooser unit 422 is selected is connected with it combines and arc A
irepresent identical word, if so, in conjunction with arc A
iin the degree of confidence currency of the word representing and arc combination, the degree of confidence of each arc is upgraded arc A
ithe degree of confidence of the word representing, triggers judgment sub-unit 425; Otherwise directly trigger judgment sub-unit 425.
Preferably, second upgrades subelement renewal arc A
irepresent the degree of confidence of word time, can be by arc A
ithe degree of confidence currency that the degree of confidence of word representing is updated to this word adds the degree of confidence minimum value of each arc in arc combination.
Adopt said method and device to determine after the degree of confidence of the word that on optimal path, each arc represents, can include but not limited to following application:
1) if the degree of confidence of certain word is lower than default confidence threshold value on optimal path, illustrate there is inaccurate recognition result in the definite recognition result of optimal path, bring poor user to experience for fear of identification error to user, can refuse to return recognition result, and can further point out user again to input voice.
2) degree of confidence of the word of determining is applied to speech recognition without in supervision adaptive technique, for the self-adaptation adjustment process of follow-up acoustic model and speech model, thereby makes speech recognition more accurate.
3) can be for recognition result is carried out to error correction, if the degree of confidence of certain word lower than default confidence threshold value, illustrates that the identification of this word exists mistake, for the error correction of recognition result provides the foundation.
The foregoing is only preferred embodiment of the present invention, in order to limit the present invention, within the spirit and principles in the present invention not all, any modification of making, be equal to replacement, improvement etc., within all should being included in the scope of protection of the invention.
Claims (10)
1. a method for definite voice identification result degree of confidence, is characterized in that, the method comprises:
S1, determine the degree of confidence of every arc in the word figure that obtains of decoding, and optimal path in definite word figure;
S2, to every arc A on described optimal path
i, in word figure, determine and this arc A
ithere is the arc set T of competitive relation;
S3, at definite described arc A
irepresent the degree of confidence of word time, from described A
iexist in the arc set T of competitive relation and determine arc A
j, wherein arc A
jwith arc A
irepresent identical word, or arc A
jbe connected arc constitutes and arc A with it
irepresent identical word; In conjunction with arc A
iwith arc A
jdegree of confidence, or further combined with described arc A
jconnect arc degree of confidence determine arc A
ithe degree of confidence of the word representing.
2. method according to claim 1, is characterized in that, in described step S1, the degree of confidence of every arc equals the value obtaining divided by the score sum in all paths in word figure through the score sum in all paths of this arc.
3. method according to claim 1, is characterized in that, determines when whether two arcs exist competitive relation, in the following ways in described step S2:
If two arc on the duration, exist overlapping, determine two arcs there is competitive relation; Or,
If two arc exists overlappingly on the duration, and the word that two arcs represent meets preset requirement in enunciative similarity, determines that two arcs exist competitive relation.
4. method according to claim 1, is characterized in that, described S3 specifically comprises:
S31, initialization arc A
ithe degree of confidence of the word representing is arc A
idegree of confidence;
S32, from described arc A
iexist in the arc set T of competitive relation and select an arc not being selected;
S33, judge the arc selected whether with arc A
irepresent identical word, if so, by arc A
ithe degree of confidence currency that the degree of confidence of word representing is updated to this word adds the degree of confidence of the arc of selection, execution step S35; Otherwise, execution step S34;
S34, judge that whether the arc that is connected with it of arc of selecting combines and arc A
irepresent identical word, if so, in conjunction with arc A
iin the degree of confidence currency of the word representing and the combination of described arc, the degree of confidence of each arc is upgraded arc A
ithe degree of confidence of the word representing, execution step S35; Otherwise directly perform step S35;
S35, judge in described arc set T whether also have non-selected arc, if so, go to described step S32; Otherwise, finish arc A
ithe degree of confidence of the word representing is determined flow process.
5. method according to claim 4, is characterized in that, described in step S34 in conjunction with arc A
iin the degree of confidence currency of the word representing and the combination of described arc, the degree of confidence of each arc is upgraded arc A
ithe degree of confidence of the word representing is specially:
By arc A
ithe degree of confidence currency that the degree of confidence of word representing is updated to this word adds the degree of confidence minimum value of each arc in the above arc combination.
6. a device for definite voice identification result degree of confidence, is characterized in that, this device comprises:
Initial determining unit, for determining the degree of confidence of every arc of word figure that decoding obtains, and optimal path in definite word figure;
Set determining unit, for every arc A on described optimal path
i, in word figure, determine and this arc A
ithere is the arc set T of competitive relation;
Degree of confidence determining unit, at definite described arc A
irepresent the degree of confidence of word time, from described A
iexist in the arc set T of competitive relation and determine arc A
j, wherein arc A
jwith arc A
irepresent identical word, or arc A
jbe connected arc constitutes and arc A with it
irepresent identical word; In conjunction with arc A
iwith arc A
jdegree of confidence, or further combined with described arc A
jconnect arc degree of confidence determine arc A
ithe degree of confidence of the word representing.
7. device according to claim 6, is characterized in that, described initial determining unit determines that the degree of confidence of every arc equals the value obtaining divided by the score sum in all paths in word figure through the score sum in all paths of this arc.
8. device according to claim 6, is characterized in that, described set determining unit is in the time that whether definite two arcs exist competitive relation, in the following ways:
If two arc on the duration, exist overlapping, determine two arcs there is competitive relation; Or,
If two arc exists overlappingly on the duration, and the word that two arcs represent meets preset requirement in enunciative similarity, determines that two arcs exist competitive relation.
9. device according to claim 6, is characterized in that, described degree of confidence determining unit specifically comprises:
Initialization subelement, for initialization arc A
ithe degree of confidence of the word representing is arc A
idegree of confidence, trigger arc chooser unit;
Arc chooser unit, for after being triggered from described arc A
iexist in the arc set T of competitive relation and select an arc not being selected;
First upgrades subelement, for judge the arc selected described arc chooser unit whether with arc A
irepresent identical word, if so, by arc A
ithe degree of confidence currency that the degree of confidence of word representing is updated to this word adds the degree of confidence of the arc of selection, triggers judgment sub-unit; Otherwise trigger second and upgrade subelement;
Second upgrades subelement, for judging that whether the arc that is connected with it of arc of selecting described arc chooser unit combines and arc A
irepresent identical word, if so, in conjunction with arc A
iin the degree of confidence currency of the word representing and the combination of described arc, the degree of confidence of each arc is upgraded arc A
ithe degree of confidence of the word representing, triggers judgment sub-unit; Otherwise directly trigger judgment sub-unit;
Judgment sub-unit, for judging whether described arc set T also exists non-selected arc, if so, triggers described arc chooser unit, otherwise finishes arc A
ithe degree of confidence of the word representing is determined flow process.
10. device according to claim 9, is characterized in that, described second upgrades subelement upgrades arc A
irepresent the degree of confidence of word time, specifically by arc A
ithe degree of confidence currency that the degree of confidence of word representing is updated to this word adds the degree of confidence minimum value of each arc in the above arc combination.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210459131.7A CN103810997B (en) | 2012-11-14 | 2012-11-14 | A kind of method and apparatus for determining voice identification result confidence level |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210459131.7A CN103810997B (en) | 2012-11-14 | 2012-11-14 | A kind of method and apparatus for determining voice identification result confidence level |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103810997A true CN103810997A (en) | 2014-05-21 |
| CN103810997B CN103810997B (en) | 2018-04-03 |
Family
ID=50707676
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210459131.7A Active CN103810997B (en) | 2012-11-14 | 2012-11-14 | A kind of method and apparatus for determining voice identification result confidence level |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103810997B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106157969A (en) * | 2015-03-24 | 2016-11-23 | 阿里巴巴集团控股有限公司 | The screening technique of a kind of voice identification result and device |
| CN109243460A (en) * | 2018-08-15 | 2019-01-18 | 浙江讯飞智能科技有限公司 | A method of automatically generating news or interrogation record based on the local dialect |
| CN111341305A (en) * | 2020-03-05 | 2020-06-26 | 苏宁云计算有限公司 | Audio data labeling method, device and system |
| CN113223500A (en) * | 2021-04-12 | 2021-08-06 | 北京百度网讯科技有限公司 | Speech recognition method, method for training speech recognition model and corresponding device |
| CN115862600A (en) * | 2023-01-10 | 2023-03-28 | 广州小鹏汽车科技有限公司 | Voice recognition method and device and vehicle |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7092883B1 (en) * | 2002-03-29 | 2006-08-15 | At&T | Generating confidence scores from word lattices |
| WO2007066297A1 (en) * | 2005-12-08 | 2007-06-14 | Koninklijke Philips Electronics N.V. | Speech recognition system with huge vocabulary |
| US20070219797A1 (en) * | 2006-03-16 | 2007-09-20 | Microsoft Corporation | Subword unit posterior probability for measuring confidence |
| CN101118745A (en) * | 2006-08-04 | 2008-02-06 | 中国科学院声学研究所 | Confidence degree quick acquiring method in speech identification system |
-
2012
- 2012-11-14 CN CN201210459131.7A patent/CN103810997B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7092883B1 (en) * | 2002-03-29 | 2006-08-15 | At&T | Generating confidence scores from word lattices |
| WO2007066297A1 (en) * | 2005-12-08 | 2007-06-14 | Koninklijke Philips Electronics N.V. | Speech recognition system with huge vocabulary |
| US20070219797A1 (en) * | 2006-03-16 | 2007-09-20 | Microsoft Corporation | Subword unit posterior probability for measuring confidence |
| CN101118745A (en) * | 2006-08-04 | 2008-02-06 | 中国科学院声学研究所 | Confidence degree quick acquiring method in speech identification system |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106157969A (en) * | 2015-03-24 | 2016-11-23 | 阿里巴巴集团控股有限公司 | The screening technique of a kind of voice identification result and device |
| CN106157969B (en) * | 2015-03-24 | 2020-04-03 | 阿里巴巴集团控股有限公司 | Method and device for screening voice recognition results |
| CN109243460A (en) * | 2018-08-15 | 2019-01-18 | 浙江讯飞智能科技有限公司 | A method of automatically generating news or interrogation record based on the local dialect |
| CN111341305A (en) * | 2020-03-05 | 2020-06-26 | 苏宁云计算有限公司 | Audio data labeling method, device and system |
| CN111341305B (en) * | 2020-03-05 | 2023-09-26 | 苏宁云计算有限公司 | Audio data labeling method, device and system |
| CN113223500A (en) * | 2021-04-12 | 2021-08-06 | 北京百度网讯科技有限公司 | Speech recognition method, method for training speech recognition model and corresponding device |
| CN113223500B (en) * | 2021-04-12 | 2022-02-25 | 北京百度网讯科技有限公司 | Speech recognition method, method for training speech recognition model and corresponding device |
| CN115862600A (en) * | 2023-01-10 | 2023-03-28 | 广州小鹏汽车科技有限公司 | Voice recognition method and device and vehicle |
| CN115862600B (en) * | 2023-01-10 | 2023-09-12 | 广州小鹏汽车科技有限公司 | Voice recognition method and device and vehicle |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103810997B (en) | 2018-04-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11322153B2 (en) | Conversation interaction method, apparatus and computer readable storage medium | |
| CN107644638B (en) | Audio recognition method, device, terminal and computer readable storage medium | |
| CN110473531B (en) | Voice recognition method, device, electronic equipment, system and storage medium | |
| CN103065630B (en) | User personalized information voice recognition method and user personalized information voice recognition system | |
| US8972260B2 (en) | Speech recognition using multiple language models | |
| US10037758B2 (en) | Device and method for understanding user intent | |
| KR20190020119A (en) | Error correction methods and devices for search terms | |
| US8666739B2 (en) | Method for estimating language model weight and system for the same | |
| JP2019504358A (en) | Speech recognition system | |
| US10242670B2 (en) | Syntactic re-ranking of potential transcriptions during automatic speech recognition | |
| WO2017166650A1 (en) | Voice recognition method and device | |
| US7392186B2 (en) | System and method for effectively implementing an optimized language model for speech recognition | |
| US7058575B2 (en) | Integrating keyword spotting with graph decoder to improve the robustness of speech recognition | |
| US8849668B2 (en) | Speech recognition apparatus and method | |
| CN103810997A (en) | Method and device for determining confidence of voice recognition result | |
| KR20120029413A (en) | Recognition using re-recognition and statistical classification | |
| CN110021293B (en) | Voice recognition method and device and readable storage medium | |
| CN103971685A (en) | Method and system for recognizing voice commands | |
| US10152298B1 (en) | Confidence estimation based on frequency | |
| CN104157285A (en) | Voice recognition method and device, and electronic equipment | |
| US8126715B2 (en) | Facilitating multimodal interaction with grammar-based speech applications | |
| CN104714954A (en) | Information searching method and system based on context understanding | |
| WO2012004955A1 (en) | Text correction method and recognition method | |
| CN111386566A (en) | Device control method, cloud device, intelligent device, computer medium and device | |
| CN105469801A (en) | Input speech restoring method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| EXSB | Decision made by sipo to initiate substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |