TW201023175A - System for consulting dictionary by speech recognition and method thereof - Google Patents

Abstract

A system for consulting a dictionary by speech recognition and method thereof is provided to solve a problem of being ineffective in querying a word which is known its pronunciation but its spelling. By recognizing speech signals of the word, analyzing the syllables produced by preceding step, and comparing the syllables with presetting samples to offer statement of the word. It solves the previous problem indeed; and further, it reduces the system overload.

Description

201023175 IX. INSTRUCTIONS: [Technical Fields of the Invention] A system for querying words by voice and a method thereof, in particular, a single-touch system and a method for providing a single voice of a voice signal. [Prior Art] Foreign language learners ^ words 'portable electronic dictionary · electronic dictionary ❹ almost indispensable learning tools; carrying (four) sub-dictionary and online electronic A, compared to each other, the former has easy to carry, The advantage of being able to query at any time, while the latter has the advantage of a large amount of data and economic benefits. But whether you use a portable electronic dictionary or an online electronic dictionary, if the user wants to query a word that only hears its pronunciation but does not know its spelling, it can only use the pronunciation to guess the spelling of the word. Enter the electronic dictionary into the light check. In the case of user guessing errors, the electronic dictionary will reply to the user to check the word, or even display another single statement that the user does not want. For example, the user originally wants to query a short-sounding word (4), but enters the long-spoken spelling peach because of misunderstanding. At this time, the electronic dictionary will have words and explanations that are not intended by the user; such a situation not only causes inconvenience of the query. Because the user takes such a trial and error query method, which will take a long time to obtain the required word explanation, this will also increase the online food inquiry system. In addition, because the keyboard of the portable electronic dictionary is small and the error rate is relatively small, the above-mentioned trial and error process will be more inconvenient. In summary, it can be seen that in the prior art, when the query only knows its pronunciation and does not know its 5 201023175 spelling words, there is always a technical means that cannot be effectively raised, and it is necessary to solve the problem. Contents] ^: The problem of the prior art that cannot be used to know the pronunciation without knowing the kind of effective explanation of the words. The present invention discloses a system for searching for words through δβ θ and a method thereof, wherein: The solution (4) system includes θ, ie, correction library, audio file database, word database, receiving module, 识2 module, search module, confirmation module and display module. Among them, the syllable database stores a plurality of syllables, the syllables of the syllables, the syllables, the syllables, the syllables, the sounds, the sounds, the sounds, the sounds, the sounds, the sounds, the sounds, the sounds, the sounds, the sounds, the sounds, the sounds, the One of the words of the test word is used to receive the voice signal; the voice recognition module is configured to perform endpoint processing on the received voice, and extract the feature of the voice signal according to the #4 of the text. According to the characteristics of the voice signal, the voice is divided into at least a syllable; the syllable database of the Chalailang finds the syllable file corresponding to each syllable, extracts the letter combination corresponding to each syllable file, and sequentially puts the letters The combination is combined into a word; the confirmation module is configured to confirm that the sound file database has a sound file keyword that matches one of the words, and further search for the word data corresponding to the sound file keyword in the word database; the display module is Used to display the word data. The method for querying a word by voice according to the present invention pre-establishes a 6 201023175 syllable database including a plurality of syllable files and a letter combination corresponding to each of the syllable files, including a plurality of sound channels and corresponding sound files. - The sound file is off, the sound file database of the word, and the keywords corresponding to each of the sound files are included. - a word library of sub-word data; a receiving-speech signal; performing endpoint detection on the voice signal; extracting a feature of the voice signal according to the sample detected by the endpoint', according to the characteristics of the voice signal, cutting the voice signal Divided into at least - syllables; find the syllable groove 'singing the syllables' in the syllable database to extract the corresponding letter combination; the letter combinations are sequentially combined into a single f 'the exact sound file database has the same Paste - the sound side of the key; in the child. The bow database searches for the word data corresponding to the sound _ word; finally, the word data is displayed. The system and method disclosed by the present invention are different from the prior art in that the present invention has a technical means for performing speech recognition on a received voice signal and analyzing and comparing the syllables generated by the voice touch; The invention can achieve the technical effect of querying words through voice and effectively providing word interpretation. [Embodiment] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings and embodiments, so that the application of the technical means to solve the technical problems and the realization of the technical effects can be fully understood and implemented. Fig. 1 is a block diagram showing a system for querying words by voice according to the present invention. Please refer to the "figure map". The voice word query system 1〇〇 includes a syllable database 11〇, a sound file database 12〇, a word database 130, a receiving module 14〇, a voice recognition module 15〇, and a search. Module 16〇, 7 201023175 Module 180. Among them, the syllable database 110 includes θ slot and corresponding syllable slot - letter combination syllable 楷

It can be ".wav", ".mp3" or other format sounds, and the corresponding combination of the child and the parent can not be ambiguous, ___ numbered letter combination ~~~~~ phonetic symbol 1 A ei syllable number f, The system seems to be "A |. The grooving beech library 120 stores a plurality of sound files. The sound should be the sound of the soundtrack. The source of the sound can be a real human voice. The hard format can be: "", "." 11) 3"' or other format sound rights; this differs from the aforementioned syllable file in the sound file of the syllable county syllable, and the sound file is the sound file of the sound file keyword. The word database m is stored with corresponding sound channels.

The key of the key is the material, and the word data corresponding to the audio key "A" is as shown in "Fig. 3". The receiving module 140 is configured to receive a voice signal recorded by the user as shown in FIG. 4, and the voice recognition module 15 is configured to perform a payment on the received message number, and extract the sample. The characteristics of the voice signal, Wei _ 音 and the county tone subtraction sign, the voice signal is divided into at least - syllables. The speech fiber technology focuses on the capture of the characteristics of the speech signal so that it can be compared to the reference waveform (syllabic file) for identification. The speech recognition module 15 will be described in detail below with regard to its possible implementation details. 8 201023175 It is reasonable to assume that the change of voice signal is continuous and slow, so the common practice is to divide the received voice signal into several short time intervals (sound frames), which can range from 20ms to 30ms, and each The window is windowed; the window is generally the most common in the Hamming Window. It has the characteristics of suppressing both ends of the frame and maintaining the middle section. Others have a rectangular window (RectangUiar window) and a Hanni window ( Hanning Window). The speech recognition module 150 can perform endpoint detection for each frame by using Short-Time Energy and Average Magnitude and Short-Time Average Zero-Crossing Rate; short-time energy Representing the amplitude, the subsequent small noise will be deleted based on the short-term energy data. The short-time zero-zero rate is the number of times the signal waveform crosses the axis, which is used as the basis for judging the length of the cycle. Assume that the sampling of the voice signal is freshly 8 Hz, the length of each __ sound box is 2Qms, and 160 sampling points are shared, and a short time ^ energy and a short time zero rate are calculated every 2 〇 ms (ie, a sound box). And Ei represents the short-term energy of the third-order frame, and Zi represents the short-term zero-crossing rate of the i-th segment. According to the above-mentioned end point system, the county's features of the voice box are reversed. Since the whole sentence of the general word pronunciation is small, the amount of data is not as large as the whole sentence. Therefore, the short-term energy and the short-term time of each sound box can be quickly sifted. Zero rate sorting (small to large), and find the median value, '1/3 of the 'and the correction', to calculate the value and the valley value in the step-by-step; where the valley value is located in the first At the lowest energy frame in the adjacent 10-segment box, if such a 201023175 valley does not exist, the valley is defined as the smallest frame between the two adjacent peaks. The voiced voice has a large amplitude and its pitch sound ^ (running the scales low) feeding point 'using the above calculated value to sing the voice signal; its towel, miscellaneous mechanical sound _ heart position, ' _ adjacent peak The first valley between them serves as a demarcation point for the syllable to divide the voice signal into at least one syllable. Then, each syllable is trimmed, and the gene week of each peak is detected to merge two adjacent peaks with a pitch of less than 80 ms, and the valley between the two peaks is deleted, and the peak of the towel energy is taken as a new peak. The peak point; in addition, for the peak of the energy too small (less than 4,) and the valley of the energy too large (greater than the soldiers), the neighboring pitch period parameters are detected to delete the unstable peak/valley, Finally, the starting point and ending point of each syllable are determined. The search module 160 obtains Linear Predictive Cepstrum Coefficients (LPCC) ® $ various parameters through linear prediction analysis to compare and analyze the syllables of the syllables and the syllable database. A syllable file that conforms to each syllable; in the case where a sub-word that is not intended by the user is displayed due to an incorrect or unclear sound recorded by the syllable, if the user wants to query "pitch", the erroneous pronunciation is [ Pjt • 显示 and the display is “Peach”, so the comparison standard can be set to 5〇% matching, and the letter combination corresponding to each syllable file (such as ba and by) is extracted, and the combination of the mother and the child are sequentially combined. Be at least one word (like baby). For each word, the confirmation module 170 is used to confirm that the sound file database 201023175 12G has a sound-over key that matches the word, and proceeds to step in the word database 130. The paste is displayed by the display module 18G. At this time, the field can be increased by the pronunciation of $(7) and the pronunciation comparison 520 (see "SA"), and the voice word query is selected. a group (not shown) for providing a user to select the standard pronunciation 51 〇 option or the pronunciation comparison 52 ,, the output module outputs a sound file corresponding to the audio key, or output the comparison sound channel It is more complicated than the voice of the ship. If the confirmation module 17 is far away, the file library 12〇 does not have any audio keywords that match the words, the words can be entered into the word 13G money and the words in the word; Confirmation in the word database 13 when there is a word data matching the word 'display the word data, and add ETTS (English Text to Speech) 53曰 (see "5B") for When the user clicks, the output mode and the group output Εττ according to the word age. Her 18G_- is told to make the time lost. "The second @" is based on (4) the method of querying the word through the voice. Please refer to "Fig. 2" to pre-establish the syllable database no, the audio file database m and the word database (9), wherein the syllable database H0 stores a plurality of syllable files and one of the corresponding syllable slots. The sound channel database 120 stores a plurality of sound files and a sound box keyword corresponding to each of the sound files, and the word database 13 stores one word data corresponding to each of the sound key keywords; receiving-voice Signal, and the end of the voice signal (dip (10)), the cans of the county, extract the 201023175 voice signal · (dip 22G); according to the characteristics of the Wei voice reduction, the voice signal is divided into at least one syllable (Step 23〇); Find the sound _ of the syllables in the syllable database no, to extract the towels each ==Γ: 240); combine the letter combinations in sequence - 2 (step 5〇), confirm the sound The file database (10) has the word ❹ ❹ 曰. The file is closed (step 26〇); in the word database 13〇, the word information corresponding to the key key is searched, and the word data is displayed (step move). And using the h-time distance i quantity and the short-time zero-rate rate for endpoint detection', and the sample generated by the endpoint detection is used as a basis for judging the voiced sound; in addition, the step of the step 230 includes the steps of trimming the syllables to The starting point and the ending point of each syllable are determined most. While the step of displaying the word data in step 270, the outputting of the z corresponding audio file key or the ETTS pronunciation may be further included according to the external operation of the user. The invention is for receiving the voice signal recorded by the user, and is divided into at least one sound by using the technique of the end point detection f and the like, and the sound is divided into at least one sound, and then transferred to each sound with a technique such as linear measurement analysis. Presetting the syllable file 'to take the letter combination corresponding to each syllable slot, and combining the letter combinations into the word 'word' and finding the phonogram keyword corresponding to the word; if there is the syllable keyword, The word data of the audio file keyword is further displayed. This (4) can be used as a master-slave architecture. ~ In summary, it can be seen that the difference between the present invention and the prior art is that there is a technical means for performing speech recognition on the received speech signal, and analyzing and comparing the speech recognition generation 12 201023175 by the technique. The means can solve the problems existing in the prior art, thereby achieving the technical effect of querying the words through the voice and effectively providing the meaning of the words. While the embodiments of the present invention are as described above, the above description is not intended to directly limit the scope of the invention. Any changes in the form and details of the embodiments may be made without departing from the spirit and scope of the invention. The scope of the invention is to be determined by the scope of the appended claims. BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a block diagram showing a system for querying words by voice according to the present invention. Fig. 2 is a flow chart showing the steps of the method for querying words by voice according to the present invention. Figure 3 is a schematic diagram of the data of the word database of the present invention. Figure 4 is a schematic diagram of the received voice signal. Figure 5A is a schematic diagram of the interface for confirming the audio file of the corresponding word data. Figure 5B is a schematic diagram of the interface for confirming the audio file without corresponding word data. [Main component symbol description] 100 voice word inquiry system 110 syllable database 13 201023175 120 audio file database 130 word database 140 receiving module 150 voice recognition module 160 search module 170 confirmation module 180 display module

510 Standard pronunciation 520 Pronunciation comparison 530 ETTS pronunciation Step 2H) Receive-speech signal and perform end point detection on the voice signal Step 220 extracts the characteristics of the voice signal based on the sample detected by the endpoint

Step 230, according to the feature of the voice subtraction, dividing the voice signal into at least one syllable step 240, searching for the sound slot corresponding to each syllable in the syllable database, and extracting the letter combination corresponding to each syllable slot. Sorting each of the letter combinations, generating a word step 260, confirming that the sound file database contains a sound file keyword 270 corresponding to the word, and searching for the word data corresponding to the sound file keyword 201023175 in the word database. Display

15

Claims (1)

201023175 ΛApplication for Patent Park·· A kind of toning sound inquiry list __, which contains: Two-tone (four) material library 'storage surface number of material files and one of the corresponding letter files of each file; slot:::= 'Stored with multiple audio files and corresponding to the sound Sub-family: a billiard library, storing the word data corresponding to each of the audio file keywords; a receiving module for receiving the voice signal; a voice discriminating group, the voice interrupting the endpoint detection, according to the endpoint Detecting the sample, extracting the feature of the voice signal, and dividing the voice signal into at least one syllable according to the characteristics of the voice signal; a search module for finding the syllable file corresponding to each syllable in the syllable database, extracting a letter combination corresponding to each syllable file and sequentially splicing each letter combination to generate a word; For confirming that the sound file database has a '§' building keyword that matches the word, and further searching for the word data corresponding to the sound file keyword in the word database; and a display module for displaying The word information. 2. The system for querying words by voice as described in claim 1 wherein the speech recognition module is configured to perform speech recognition based on voiced features. 201023175 * The system for querying words by voice in the first paragraph of claim 1 of the patent scope is as follows: after the segmentation of the speech is reduced to at least - _ syllables, it is further included for each syllable trim. • The system for querying words by voice as described in claim 1 of the patent scope, the system further includes an output module that outputs a sound file corresponding to the sound box keyword. A system for querying a word by voice as described in item i of the scope of the patent application, wherein the generation of the sound file can be a human voice or an ETTS pronunciation. 6. If you apply for a patent range! The voice word query system described in the section can be extended to the master-slave architecture. '~ kind of transcendence query one-touch method, the built-in snake contains a plurality of sounds and files corresponding to each syllable file-letter combination _ syllable data 2 plural sound files and one of the audio file keys corresponding to each of the sound files a file database, and a sub-word database containing a word corresponding to each of the key words, the method comprising the following steps: receiving a voice signal; performing an endpoint on the voice signal Detecting, according to the sample detected by the endpoint, extracting the feature of the voice signal; and according to the characteristics of the voice signal, dividing the voice signal into at least one syllable; finding the syllable corresponding to each syllable in the syllable database The slot 'extracts the letter corresponding to each of the syllable files; 17 201023175 sequentially combines the letter combinations to generate a word; confirm that the sound file database contains a phonetic keyword that matches one of the words; 3 S The word database finds the word data corresponding to the phonetic keyword; and displays the word data. 8. e 9. 10. 11. For the method of querying words by voice as described in item 7 of the patent application, the sample of the endpoint detection is used as the basis for judging the voiced sound. For example, if the application for the vocalization of the vocalization is described in the seventh section of the syllabus, the voice signal is divided into at least _ syllables, and the syllables are trimmed. step. For example, the method for transacting a voice through a word as described in claim 7 of the patent scope further includes outputting a sound file corresponding to the audio file keyword. For example, the method of applying for the word-recognition by the voice of the seventh section is applied. The voice file can be generated by a real person or by a sound. This method can be extended to the master-slave architecture by applying the method of voice-resolving words in the 7th remuneration. 12.