ZHANG, Shi, Lei (IBM China, A#1-3/F B#1-3/F C#1-3/F, Diamond, Zhong Guang Cun Software Park, Haidian District, Beijing 11 3, 10019, CN)
SHUANG, Zhiwei (IBM China, A#1-3/F B#1-3/F, C#1-3/F, Diamond, Zhong Guang Cun Software Park, Haidian District, Beijing 11 3, 10019, CN)
QIN, Yong (IBM China, A#1-3/F B#1-3/F, C#1-3/F, Diamond, Zhong Guang Cun Software Park, Haidian District, Beijing 11 3, 10019, CN)
IBM UNITED KINGDOM LIMITED (P.O. Box 41, North Harbour, Portsmouth Hampshire PO6 3AU, GB)
SHI, Qin (IBM China, A#1-3/F B#1-3/F, C#1-3/F, Diamond, Zhong Guang Cun Software Park, Haidian District, Beijing 93, 10010, CN)
ZHANG, Shi, Lei (IBM China, A#1-3/F B#1-3/F C#1-3/F, Diamond, Zhong Guang Cun Software Park, Haidian District, Beijing 11 3, 10019, CN)
SHUANG, Zhiwei (IBM China, A#1-3/F B#1-3/F, C#1-3/F, Diamond, Zhong Guang Cun Software Park, Haidian District, Beijing 11 3, 10019, CN)
QIN, Yong (IBM China, A#1-3/F B#1-3/F, C#1-3/F, Diamond, Zhong Guang Cun Software Park, Haidian District, Beijing 11 3, 10019, CN)
| CLAIMS 1. A method for assessing speech prosody, comprising: receiving input speech data; acquiring prosody constraint having a rhythm feature constraint by acquiring a standard rhythm feature corresponding to the input speech data; assessing prosody of the input speech data according to the prosody constraint by comparing a rhythm feature of the input speech data with the corresponding standard rhythm feature; and providing assessment result. 2. The method according to claim 1, wherein the rhythm feature is represented as phrase boundary location, the phrase boundary comprises at least one of the following: silence and pitch reset. 3. The method according to claim 2, wherein the step of comparing rhythm feature of the input speech data with corresponding standard rhythm feature further comprises: checking whether phrase boundary location of the input speech data matches with phrase boundary location of the standard rhythm feature. 4. The method according to claim 2, wherein the step of acquiring rhythm feature of the input speech data further comprises: acquiring input text data corresponding to the input speech data; aligning the input text data with the input speech data; and measuring phrase boundary location of the input speech data. 5. The method according to claim 4, wherein the step of acquiring standard rhythm feature corresponding to the input speech data further comprises: processing the input text data to acquire a corresponding input language structure; matching the input language structure with a standard language structure of standard speech in a standard corpus to determine occurrence probability of phrase boundary location of the input text data; and extracting phrase boundary location of the standard rhythm feature. 6. The method according to claim 5, wherein the step of extracting phrase boundary location of the standard rhythm feature further comprises: extracting phrase boundary location whose occurrence probability is above a certain threshold. 7. The method according to claim 5, where the step of matching the input language structure with a standard language structure of standard speech in a standard corpus to determine occurrence probability of phrase boundary location of the input text data comprises: traversing a decision tree of the standard language structure according to input language structure of at least one word of the input text data to determine occurrence probability of phrase boundary location of the at least one word. 8. The method according to any one of the preceding claims, wherein the prosody constraint comprises fluency feature constraint, the method further comprising: acquiring fluency feature of the input speech data. 9. The method according to claim 8, wherein the step of acquiring fluency feature of the input speech data further comprises: acquiring input text data corresponding to the input speech data; aligning the input text data with the input speech data; and measuring fluency feature of the input speech data. 10. The method according to claim 8, wherein the fluency feature comprises the total number of phrase boundaries within one sentence, and the phrase boundary comprises at least one of the following: silence and pitch reset, the step of acquiring prosody constraint further comprising: determining a predicted value of the total number of phrase boundaries according to sentence length of the text data corresponding to the input speech data, the step of assessing prosody of the input speech data according to the prosody constraint further comprising: comparing the total number of phrase boundaries of the input speech data with the predicted value of the total number of phrase boundaries. 11. The method according to claim 8, wherein the fluency feature comprises silence duration of phrase boundary, the step of acquiring prosody constraint further comprising: acquiring standard silence duration corresponding to the input speech data, and the step of assessing prosody of the input speech data according to the prosody constraint further comprising: comparing the silence duration of phrase boundary of the input speech data with the corresponding standard silence duration. 12. The method according to claim 11, wherein the step of acquiring standard silence duration corresponding to the input speech data further comprises: processing the input text data to obtain a corresponding input language structure; and matching the input language structure with a standard language structure of standard speech in a standard corpus to determine standard silence duration of phrase boundary of the input text data. 13. The method according to claim 12, wherein the step of matching the input language structure with a standard language structure of standard speech in a standard corpus to determine standard silence duration of phrase boundary of the input text data comprises: traversing a decision tree of the standard language structure according to input language structure of at least one word of the input text data to determine standard silence duration of phrase boundary of the at least one word, wherein the standard silence duration is an average value of the silence duration of phrase boundary of standard language structures for which statistics have been gathered. 14. The method according to claim 8, wherein the fluency feature comprises number of repetition times of a word, the step of acquiring prosody constraint further comprising: acquiring a permissible value of the number of repetition times; and the step of assessing prosody of the input speech data according to the prosody constraint further comprising: comparing the number of repetition times of the input speech data with the permissible value. 15. The method according to claim 8, wherein the fluency feature comprises phone hesitation degree, the phone hesitation degree including at least one of number of phone hesitation times or phone hesitation duration, the step of acquiring prosody constraint further comprising: acquiring a permissible value of the phone hesitation degree; and the step of assessing prosody of the input speech data according to the prosody constraint further comprising: comparing the phone hesitation degree of the input speech data with the permissible value of the phone hesitation degree. 16. The method according to any one of the preceding claims, further comprising: adding the input speech data into the corpus as standard speech data according to the assessment result. 17. The method according to any one of the preceding claims, wherein the assessment result comprises at least one of the following: score of prosody of the input speech data; detailed analysis on prosody of the input speech data; reference speech data. 18. A system for assessing speech prosody, comprising: an input speech data receiver configured for receiving input speech data; a prosody constraint acquiring means configured for acquiring prosody constraint; an assessing means configured for assessing prosody of the input speech data according to the prosody constraint; and a result providing means configured for providing assessment result. 19. The system according to claim 18, wherein the prosody constraint comprises rhythm feature constraint. 20. The system according to claim 19, further comprising: a rhythm feature acquiring means configured for acquiring rhythm feature of the input speech data, the rhythm feature is represented as phrase boundary location, the phrase boundary comprises at least one of the following: silence and pitch reset; and the prosody constraint acquiring means is further configured for acquiring standard rhythm feature corresponding to the input speech data, the assessing means is further configured for comparing rhythm feature of the input speech data with corresponding standard rhythm feature. 21. The system according to claim 18, wherein the prosody constraint comprises fluency feature constraint, the system further comprising: a fluency feature acquiring means configured for acquiring fluency feature of the input speech data, the prosody feature acquiring means is further configured for: acquiring input text data corresponding to the input speech data; aligning the input text data with the input speech data; and measuring fluency feature of the input speech data. 22. The system according to claim 23, wherein the fluency feature comprises at least one of the following: total number of phrase boundaries, the phrase boundary comprises at least one of silence or pitch reset; silence duration of the phrase boundary; number of repetition times of a word; and phone hesitation degree, the phone hesitation degree including at least one of number of phone hesitation times or phone hesitation duration. 23. A system for assessing speech prosody comprising corresponding means for implementing steps in any one of method claims 1 to 17. |
Technical Field of the Invention
This invention generally relates to a method and system for assessing speech, in particular, to a method and system for assessing prosody of speech data.
Background of the Invention
Speech assessment is an important area in speech application technology, the main purpose of which is to assess the quality of input speech data. However, speech assessment technologies in the prior art mainly focus on assessing pronunciation of input speech data, namely, distinguishing and scoring pronunciation variance of speech data. Take the word "today" for example, the correct American pronunciation should be [ts'de], whereas a reader might mispronounce it as [tu'de]. The existing speech assessment technologies may detect and correct incorrect pronunciations. If the input speech data is a sentence or a long paragraph rather than a word, the sentence or paragraph needs to be segmented first so as to perform force alignment between the input speech data and corresponding text data, and then an assessment is performed according to pronunciation variance of each word. In addition, most of the existing speech assessment products require a reader to read given speech information, which includes read text of some paragraph or read after a piece of standard speech, such that the input speech data is restricted by given content.
US patent publication US 2009/0204398 Al "Measurement of Spoken Language Training, Learning & Testing" discloses how fluency of a spoken utterance or passage is measured and presented to the speaker and to others. A method is described that includes recording a spoken utterance, evaluating the spoken utterance for accuracy, evaluating the spoken utterance for duration, and assigning a score to the spoken utterance based on the accuracy and the duration. Summary of the Invention
The inventor of the present invention has noticed that the prior art fails to provide an effective method and system for assessing speech prosody. Furthermore, a majority of the prior arts require readers to follow reading of given text/speech, which limits the application scope of prosody assessment. The present invention sets forth an effective method and system for assessing input speech. Further, the invention does not have any restriction on input speech data, that is, user can read certain text/speech or follow reading thereof, and the user can also give a free speech. Therefore, the present invention not only can assess prosody of a reader or follower, but also can assess prosody of any piece of input speech data. The present invention not only can help a self- learner to score and correct his own spoken language, but also can assist an examiner to assess an examinee's performance during an oral test. The present invention not only can be implemented as a special hardware device such as repeater, but also can be implemented as software logic in a computer to operate in conjunction with a sound collecting device. The present invention not only can serve one end user, but also can be adopted by a network service provider so as to assess input speech data of a plurality of end users.
In particular, the present invention provides a method for assessing speech prosody, comprising: receiving input speech data; acquiring prosody constraint; assessing prosody of the input speech data according to the prosody constraint; and providing assessment result. In embodiments, the present invention employs a prosody structure prediction model, to generate boundary probability for words. A comparison of the boundary probability with a detected boundary, obtained from speech recognition, allows for an evaluation of speech rhythm. Further, the detection and evaluation of speech hesitation and/or fluency is achieved by comparing the phone duration with a pre-defined duration.
The present invention also provides a system for assessing speech prosody, comprising: an input speech data receiver for receiving input speech data; a prosody constraint acquiring means for acquiring prosody constraint; an assessing means for assessing prosody of the input speech data according to the prosody constraint; and a result providing means for providing assessment result. Brief Description of the Drawings
Preferred embodiments of the present invention will now be described, by way of example only, with reference to the following drawings in which:
Fig. l shows a flow chart of a method for assessing speech prosody;
Fig.2 shows a flow chart of a method for assessing rhythm according to one embodiment of the invention;
Fig.3 shows a flow chart of acquiring rhythm feature of input speech data according to one embodiment of the invention;
Fig.4 shows a flow chart of acquiring standard rhythm feature according to one embodiment of the invention;
Fig.5 shows a diagram of a portion of decision tree according to one embodiment of the invention;
Fig.6A shows a speech analysis chart of measuring silence of input speech data according to one embodiment of the invention;
Fig.6B shows a speech analysis chart of measuring pitch reset of input speech data according to one embodiment of the invention;
Fig.7 shows a flow chart of a method for assessing fluency according to one embodiment of the invention;
Fig.8 shows a flow chart of acquiring fluency feature of input speech data according to one embodiment of the invention;
Fig.9 shows a flow chart of a method for assessing total number of phrase boundaries according to one embodiment of the invention;
Fig.10 shows a flow chart of a method for assessing silence duration according to one embodiment of the invention;
Fig. l 1 shows a flow chart of a method for assessing number of repetition times of a word according to one embodiment of the invention;
Fig.12 shows a flow chart of a method for assessing phone hesitation degree according to one embodiment of the invention;
Fig.13 shows a block diagram of a system for assessing speech prosody; and
Fig.14 shows a diagram of performing speech prosody assessment in manner of network service according to one embodiment of the invention. Detailed Description of the Preferred Embodiments
In the following discussion, a large amount of specific details are provided to facilitate to understand the invention thoroughly. However, for those skilled in the art, it is evident that it does not affect the understanding of the invention without these specific details. And it will be recognized that, the usage of any of following specific terms is just for convenience of description, thus the invention should not be limited to any specific application that is identified and/or implied by such terms. The present invention sets forth an effective method and system for assessing input speech.
Further, the invention does not have any restriction on input speech data, that is, user can read certain text/speech or follow reading thereof, and the user can also give a free speech. Therefore, the present invention not only can assess prosody of a reader or follower, but also can assess prosody of any piece of input speech data. The present invention not only can help a self- learner to score and correct his own spoken language, but also can assist an examiner to assess an examinee's performance during an oral test. The present invention not only can be implemented as a special hardware device such as repeater, but also can be implemented as software logic in a computer to operate in conjunction with a sound collecting device. The present invention not only can serve one end user, but also can be adopted by a network service provider so as to assess input speech data of a plurality of end users.
Fig.1 shows a flow chart of a method for assessing speech prosody. First, at step 102, input speed data is received, for example, a sentence said by a user "Is it very easy for you to stay healthy in England". At step 104, prosody constraint is acquired, the prosody constraint may be rhythm constraint, fluency constraint or both, more details thereof are given hereinafter. At step 106, assessment is performed on prosody of the input speech data according to the prosody constraint, and assessment result is provided at step 108. Fig.2 shows a flow chart of a method for assessing rhythm according to one embodiment of the invention. First, at step 202, the input speech data is received. Then, at step 204, rhythm feature of the input speech data is acquired, the rhythm feature may be represented as phrase boundary location, the phrase boundary comprises at least one of the following: silence and pitch reset. Silence refers to a time interval between words in the speech data. Referring to Fig.6A, it shows a speech analysis chart of measuring silence of input speech data according to one embodiment of the invention. Upper portion 602 of Fig.6A is an energy curve varying with time that reveals a speaker's speech energy in unit of decibel. It can be clearly seen from Fig.6A that, the speaker is silence for 0.463590 seconds between "easy" and "for". Pitch reset refers to pitch variation between words in speech data. Usually, pitch reset may occur if the speaker needs to take a breath after finishing a word or raise pitch of the following word. Referring to Fig.6B, it shows a speech analysis chart of measuring pitch reset of input speech data according to one embodiment of the invention. Upper portion 606 of Fig.6B is an energy curve varying with time that reveals a speaker's speech energy, and the pitch variation contour shown in lower portion 608 of Fig.6B can be derived from the energy curve. A pitch reset may be identified from the pitch variation contour. Analyzing speech data to obtain the energy curve and pitch variation contour belongs to prior art, the description of which will be omitted here. It can be known form the pitch variation contour shown at 608 that, although there is no silence between word "easy" and "for", there is a pitch reset between "easy" and "for".
For a speaker, if there is no silence or pitch reset at correct location, his reading or spoken language will not be standard or native, for example, if the speaker pauses after "very" rather than "easy", as shown in the following example:
Is it very (silence) easy for you to stay healthy in England.
Apparently, if the speaker speaks in the above way, it does not conform to normal speech rhythm. The following steps are used to judge whether a speaker pauses or makes a pitch reset at a correct location.
Fig.3 shows a flow chart of acquiring rhythm feature of input speech data according to one embodiment of the invention. At step 302, input text data corresponding to the input speech data is acquired (for example, text content of "Is it very easy for you to stay healthy in England" is acquired). The conversion of speech data into corresponding text data may be performed by using any known or unknown convention technologies, the description of which will be omitted here. At step 304, the input text data is aligned with the input speech data, that is, each word in the speech data is made to correspond in time to each word in the text data. The purpose of alignment is to further analyze rhythm feature of the input speech data. At step 306, phrase boundary location of the input speech data is measured, it may be to measure after which word the speaker pauses or makes a pitch reset. Further, phrase boundary location may be marked on the aligned text data, for example:
Is it very easy (silence) for you to stay healthy in England.
Back to Fig.2, at step 206, standard rhythm feature corresponding to the input speech data is acquired. The so-called standard rhythm feature refers to that, for speech data spoken by a speaker, where should silence or pitch reset be made under standard pronunciation; or alternatively, if a professional announcer reads the same sentence, where his/her phrase boundary location should be set. Of course, for a sentence, there may be various standard phrase boundaries. For, example, the following listed probabilities may all be considered as correct or standard reading manner:
Is it very easy (silence) for you to stay healthy in England.
Is it very easy for you to stay healthy (silence) in England.
Is it very easy for you to stay healthy in England (there is no silence or pitch reset in the whole sentence).
The present invention is not only limited to assess a speaker's input speech data according to one standard reading manner; rather, it can perform assessment by comprehensively considering various standard reading manners. Details about the step of acquiring standard rhythm feature will be given below.
Fig.4 shows a flow chart of acquiring standard rhythm feature according to one embodiment of the invention. At step 402, the input text data is processed to acquire corresponding input language structure. Further, each word in the input text data may be analyzed to acquire its language structure so as to generate a language structure table of the whole sentence. Table 1 shows an example of the language structure table:
Table 1
Since standard speech data stored in a corpus are limited (such as tens of thousands of sentences or hundreds of thousands of sentences), it is difficult to find therein a sentence whose language structure is exactly the same as that of the speaker's input speech data. For example, it is difficult to find a standard speech whose language structure is also "aux pro adv adj prep pro prep vi noun prep noun". The inventor of the present invention has noticed that, although grammatical structure of the whole sentence may not be the same, similar phrase boundary may exist if grammatical structure within a certain range is the same. For instance, if a standard speech data stored in the corpus is:
Vitamin c is extremely good (silence) for all types of skin.
The above sentence also has the grammatical structure of "extremely (adv) good (adj) for (prep)", thus, phrase boundary location that should exist in the input speech data can be deduced from phrase boundaries of standard speech with similar grammatical structure. Of course, the corpus may include therein numerous standard speech data with language structure "adv adj prep", some of them have silence/pitch reset after adj; while others do not have silence/pitch reset after adj. The present invention judges whether silence/pitch reset should occur after a word based on statistic probability of phrase boundary of numerous standard speech data with identical language structure.
Specifically, at step 404, the input language structure is matched with standard language structure of standard speech in a standard corpus to determine occurrence probability of phrase boundary location of the input text data. Step 404 further comprises traversing a decision tree of the standard language structure according to input language structure of at least one word of the input text data (for instance, language structure of "easy" is "adv adj prep") to determine occurrence probability of phrase boundary location of the at least one word. The decision tree refers to a tree structure obtained from analyzing language structure of standard speech in the corpus. Fig.5 shows a diagram of a portion of decision tree according to one embodiment of the invention. According to the embodiment in Fig.5, when building a decision tree based on numerous standard speech data, it is first judged whether part of speech of current word is Adj, if the result is Yes, then it is further judged whether part of speech of its left adjacent word is Adv; if the result is No, it is judged whether part of speech of current word is Aux. If part of speech of left adjacent word is Adv, then it is further judged whether part of speech of right adjacent word is Prep; otherwise, continue to judge whether part of speech of left adjacent word is Ng. If part of speech of right adjacent word is Prep, then statistics about whether silence/pitch reset occurs after a word whose part of speech is Adj is gathered and recorded, otherwise, continue to perform other judgment on part of speech of right adjacent word. After analyzing all standard speeches in the corpus, statistics of leaf nodes are calculated so as to obtain occurrence probability of phrase boundary. For example, in standard speech data, silence/pitch reset occurs in 875 words with language structure "adv adj prep", and silence/pitch reset does not occur in 125 words with language structure "adv adj prep", then occurrence probability of phrase boundary location is 0.875000. Details about the process of building a decision tree may be further found in reference document Shi, Qin/Jiang, DanNing/Meng,FanPing/Qin,Yong (2007): "Combining length distribution model with decision tree in prosodic phrase prediction", In
INTERSPEECH-2007, 454-457. It can be seen that, by traversing the decision tree according to language structure of certain word in the input text data, occurrence probability of phrase boundary location of that word may be determined, so that occurrence probability of phrase boundary location of each word in the input speech data may further be obtained, for example:
Is(O.OOOOOO) it(0.300000) very(0.028571) easy(0.875000) for(O.OOOOOO) you(0.470588) to(0.000000) stay(0.026316) healthy(0.633333) in(0.0513514)
England( 1.000000)
At step 406, phrase boundary location of the standard rhythm feature is extracted, and phrase boundary location whose occurrence probability is above a certain threshold is further extracted. For example, if the threshold is set at 0.600000, then word whose occurrence probability of phrase boundary location is above 0.600000 will be extracted. According to the above example, "easy", "healthy" and "England" will all be extracted. That is, if in the input speech data, silence/pitch reset occurs after "England", or silence/pitch reset occurs after any one of or both of "easy" and "healthy", they may all be considered as reasonable in rhythm.
It is to be noted that, the foregoing merely gives a simple example of language structure table, actually, the language structure table may be further expanded to further comprise other items, such as: whether current word is at beginning, at end or in middle of a sentence, part of speech of a second word from its left, part of speech of a second word from its right, etc.
Back to Fig.2, at step 208, the rhythm feature of the input speech data is compared with corresponding standard rhythm feature, so as to check whether the phrase boundary location of the input speech data matches with the phrase boundary location of the standard rhythm feature, comprising: whether a speaker pauses/makes a pitch reset at a location where pause/ pitch reset should not be made, or whether a speaker does not pause/make a pitch reset at a location where pause/ pitch reset should be made. Finally, at step 210, an assessment result is provided. According to the embodiment shown in Fig.5A, the speaker pauses after "easy" and "England", so it conforms to standard rhythm feature. It is not necessary for the speaker to pause after each word whose occurrence probability of phrase boundary is above
0.600000, because this may cause too many pause times in a sentence, so that coherence of the whole sentence will be affected. The present invention may adopt various predetermined assessing strategies to perform assessment based on the comparison between rhythm feature of the input speech data and corresponding standard rhythm feature.
As mentioned above, prosody may refer to rhythm of speech data, or fluency of speech data or both. The foregoing specifically describes the method for assessing input speech data in terms of rhythm feature. The following will describe a method for assessing input speech data in terms of fluency feature.
Fig.7 shows a flow chart of a method for assessing fluency according to one embodiment of the invention. Input speech data is received at step 702. Fluency feature of the input speech data is obtained at step 704. The fluency feature comprises one or more of the following: total number of phrase boundaries within a sentence, silence duration of phrase boundary, number of repetition times of a word, phone hesitation degree. Fluency constraint is obtained at step 706, the input speech data is assessed according to the fluency constraint at step 708, and assessment result is provided at step 710.
Fig.8 shows a flow chart of acquiring fluency feature of input speech data according to one embodiment of the invention. At step 802, input text data corresponding to the input speech data is first acquired, then at step 804, the input text data is aligned with the input speech data. Steps 802 and 804 are similar to steps 302 and 304 in Fig.3, the description of which will be omitted. At step 806, fluency feature of the input speech data is measured.
Fig.9 shows a flow chart of a method for assessing total number of phrase boundaries according to one embodiment of the invention. At step 902, input speech data is first received, and then at step 904, total number of phrase boundaries of the input speech data is acquired. As mentioned above, phase boundary location of a plurality of standard rhythm features may be extracted by analyzing a decision tree. However, if pause/pitch reset is made at every phrase boundary location, fluency of the whole sentence may be affected. Thus, total number of phrase boundaries in one sentence needs to be assessed. If a speaker speaks a long paragraph of words, how to detect end of a sentence belongs to prior art and the description of which will be omitted here. At step 906, a predicted value of the total number of phrase boundaries is determined according to sentence length of text data corresponding to the input speech data. In the example listed above, the whole sentence comprises 11 words. For example, if a predicted value of total number of phrase boundaries of the sentence determined based on certain empiric value is 2, then in addition to one pause that should be made at end of the sentence, the speaker is allowed to make one pause/pitch reset at most in the middle of the sentence. At step 908, the total number of phrase boundaries of the input speech data is compared with the predicted value of the total number of phrase boundaries. At step 910, an assessment result is provided. If the speaker is silent as follows:
Is it very easy (silence) for you to stay healthy (silence) in England (silence) then although assessment result of his/her rhythm feature may be good, assessment result of fluency feature may have problem.
Fig.10 shows a flow chart of a method for assessing silence duration according to one embodiment of the invention. At step 1002, input speech data is received, and at step 1004, silence duration of phrase boundary of the input speech data is acquired, for example, silence duration after "easy" in Fig. 5A is 0.463590 second. At step 1006, standard silence duration corresponding to the input speech data is acquired. Step 1006 further comprises: processing the input text data to obtain a corresponding input language structure; and matching the input language structure with a standard language structure of standard speech in a standard corpus to determine standard silence duration of phrase boundary of the input text data. The method for acquiring input language structure has been described in detail hereinabove and the description of which will be omitted here. The step of determining standard silence duration further comprises: traversing a decision tree of the standard language structure according to input language structure of at least one word of the input text data to determine standard silence duration of phrase boundary of the at least one word, wherein the standard silence duration is an average value of the silence duration of phrase boundary of standard language structures for which statistics have been gathered. Take the decision tree in Fig. 5 for example, when building the decision tree, not only statistics about occurrence probability of phrase boundary of every word of the standard speech data in the corpus are gathered, but also statistics about silence duration are gathered so as to record average value of silence duration. For example, average silence duration of phrase boundary of "adj" in language structure "adv adj prep" is 0.30 second, thus, 0.30 second is the standard silence duration of the language structure "adv adj prep". At step 1008, silence duration of phrase boundary of the input speech data is compared with the corresponding standard silence duration, and assessment result is provided at step 1010 based on a predetermined assessing strategy. For example, the predetermined assessing strategy may be: when actual silence duration significantly exceeds standard silence duration, score of assessment result will be reduced. At step 1010, an assessment result is provided.
Fig.11 shows a flow chart of a method for assessing number of repetition times of a word according to one embodiment of the invention. At step 1102, input speech data is received, and at step 1104, number of repetition times of a word in the input speech data is acquired, for instance, a person who has impediment in speech usually will have problem in fluency. Therefore, his language fluency can be assessed according to number of repetition times of a word or phrase within one sentence or one paragraph. The number of repetition times in the present invention refers to repetition results from lack of fluency in speech; it does not include repetition intentionally made by the speaker to emphasize certain word or phrase. Repetition due to lack of fluency differs from repetition for emphasis in speech feature, the former usually will not have pitch reset during repetition, while the latter often has pitch reset accompanied therewith. For example, in the above example, if the input speech data is:
Is it very very easy for you to stay healthy in England,
that is, no pitch reset occurs between the two "very", then the repetition of "very" may be caused by lack of fluency.
If the input speech data is:
Is it very (pitch reset) very easy for you to stay healthy in England.
Then, the repetition of "very" may be caused by an emphasis intentionally made by the speaker.
At step 1106, a permissible value of the number of repetition times is acquired (for example, a word or phrase may be repeated once in a paragraph at most); and at step 1108, the number of repetition times of the input speech data is compared with the permissible value; and finally at step 1110, an assessment result of the comparison is provided.
Fig.12 shows a flow chart of a method for assessing phone hesitation degree according to one embodiment of the invention. At step 1202, input speech data is received. At step 1204, phone hesitation degree of the input speech data is acquired, the phone hesitation degree comprising at least one of number of phone hesitation times or phone hesitation duration. For example, if a speaker prolongs short vowel [i] of word "easy", it may affect his oral/reading fluency. At step 1206, a permissible value of the phone hesitation degree is acquired (for example, the maximum number of phone hesitation times or the maximum phone hesitation duration allowed within one paragraph or sentence). Then at step 1208, the phone hesitation degree of the input speech data is compared with the permissible value of the phone hesitation degree. Finally at step 1210, an assessment result of the comparison is provided.
Fig.13 shows a block diagram of a system for assessing speech prosody. The system comprises an input speech data receiver, a prosody constraint acquiring means, an assessing means, and a result providing means, wherein the input speech data receiver is for receiving input speech data, the prosody constraint acquiring means is for acquiring prosody constraint, the assessing means is for assessing prosody of the input speech data according to the prosody constraint, and the result providing means is for providing assessment result. The prosody constraint comprises one or more of rhythm constraint or fluency constraint. The system may further comprise: a rhythm feature acquiring means (not shown in the figure) for acquiring rhythm feature of the input speech data. The rhythm feature is represented as phrase boundary location, the phrase boundary comprises at least one of the following: silence and pitch reset. In addition, the prosody constraint acquiring means is further used for acquiring standard rhythm feature corresponding to the input speech data. The assessing means is further used for comparing the rhythm feature of the input speech data with the corresponding standard rhythm feature.
According to another embodiment of the present invention, the system further comprises: a fluency feature acquiring means (not shown in the figure) for acquiring fluency feature of the input speech data, and the prosodic feature acquiring means is further used for acquiring input text data corresponding to the input speech data, aligning the input text data with the input speech data, and measuring fluency feature of the input speech data.
Other functions performed by the system for assessing speech prosody shown in Fig.13 corresponds to respective steps in the method for assessing speech prosody as described above, the description of which will be omitted here.
It is to be noted that, the present invention may only assess one or more rhythm features of the input speech data, or may only assess one or more fluency features or may perform a comprehensive prosody assessment by combining one or more rhythm features and one or more fluency features. If there is more than one assessed item, different or same weights may be set for different assessed items, namely, different assessment strategies may be established based on actual need.
Although the present invention provides a method and system for assessing speech prosody, it may also be combined with other method and system for assessing speech. For instance, the system of the present invention may be combined with other speech assessing system such as a system for assessing pronunciation and/or a system for assessing grammar, etc., so as to perform a comprehensive assessment on input speech data. The result of prosody assessment of the present invention may be taken as one item of the comprehensive speech assessment and assign certain weight thereto.
According to one embodiment of the invention, based on the assessment result, the input speech data with a high score, for example, may be added into the corpus as standard speech data, thereby further enriching quantity of standard speech data.
Fig.14 shows a diagram of performing speech prosody assessment in manner of network service according to one embodiment of the invention. A server 1402 provides service of assessing speech prosody, different users may upload their speech data to the server 1402 through a network 1404, and the server 1402 may return result of prosody assessment to the user. According to another embodiment of the present invention, the system for assessing speech prosody may also be applied in a local computer for a speaker to perform speech prosody assessment. According to yet another embodiment of the present invention, the system for assessing speech prosody may also be designed as a special hardware device for a speaker to perform speech prosody assessment.
The assessment result of the present invention comprises at least one of the following: score of prosody of the input speech data; detailed analysis on prosody of the input speech data; reference speech data. The score may be assessed using a hundred-point system, five-point system or any other system; or descriptive score may be used, such as excellent, good, fine, bad, etc. The detailed analysis may comprise one or more of the following: location where speaker's silence/pitch reset is inappropriate, total number of speaker's silence/pitch reset is too high, speaker's silence duration at certain location is too long, speaker's number of repetition times of some word/phrase is too high, and speaker's phone hesitation degree of some word is too high. Meanwhile, the assessment result may also provide speech data for reference, for example, a correct way for reading the sentence "Is it very easy for you to stay healthy in England". There may be multiple pieces of reference speech data. The system of the present invention may provide one piece of reference speech data, or provide multiple pieces of speech data for reference.
Although the description above takes one English sentence as an example, the present invention has no limitation on the type of language to be assessed. The present invention may be applied to assess prosody of speech data of various languages such as Chinese, Japanese, Korean, etc.
Although the description above takes speech as an example, the present invention may also assess prosody of other phonetic forms such as singing or rap.
As will be appreciated by one skilled in the art, the present invention may be embodied as a system, method or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment (including firmware, resident software, micro-code, etc.) or an embodiment combining software and hardware aspects that may all generally be referred to herein as a "circuit," "module" or "system." Furthermore, the present invention may take the form of a computer program product embodied in any tangible medium of expression having computer usable program code embodied in the medium.
Any combination of one or more computer usable or computer readable medium(s) may be utilized. The computer-usable or computer-readable medium may be, for example but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, device, or propagation medium. More specific examples (a non- exhaustive list) of the computer-readable medium would include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CDROM), an optical storage device, a transmission media such as those supporting the Internet or an intranet, or a magnetic storage device. Note that the computer- usable or computer-readable medium could even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, via, for instance, optical scanning of the paper or other medium, then compiled, interpreted, or otherwise processed in a suitable manner, if necessary, and then stored in a computer memory. In the context of this document, a computer-usable or computer-readable medium may be any medium that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The computer-usable medium may include a propagated data signal with the computer-usable program code embodied therewith, either in baseband or as part of a carrier wave. The computer usable program code may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc.
Computer program code for carrying out operations of the present invention may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet Service Provider).
The present invention is described below with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable medium that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable medium produce an article of manufacture including instruction means which implement the function/act specified in the flowchart and/or block diagram block or blocks.
The computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks..
The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems that perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises", when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
The corresponding structures, materials, acts, and equivalents of all means or step plus function elements in the claims below are intended to include any structure, material, or act for performing the function in combination with other claimed elements as specifically claimed. The description of the present invention has been presented for purposes of illustration and description, but is not intended to be exhaustive or limited to the invention in the form disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope of the invention. The embodiment was chosen and described in order to best explain the principles of the invention and the practical application, and to enable others of ordinary skill in the art to understand the invention for various embodiments with various modifications as are suited to the particular use contemplated.
Next Patent: GENE TRAPPING FOR USE IN GENE THERAPY