Classifications

• • G—PHYSICS
  • G10—MUSICAL INSTRUMENTS; ACOUSTICS
  • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
  • G10L15/00—Speech recognition
  • G10L15/22—Procedures used during a speech recognition process, e.g. man-machine dialogue
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F40/00—Handling natural language data
  • G06F40/10—Text processing
  • G06F40/166—Editing, e.g. inserting or deleting
• • G—PHYSICS
  • G10—MUSICAL INSTRUMENTS; ACOUSTICS
  • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
  • G10L15/00—Speech recognition
  • G10L15/22—Procedures used during a speech recognition process, e.g. man-machine dialogue
  • G10L2015/223—Execution procedure of a spoken command

Definitions

• the invention relates to a system and method for reviewing a text recognized by a speech recognizer.
• US 5,031,113 describes a speech recognition system used for dictation.
• the system enables a user to review the recognized text in a so-called synchronous reproduction mode. After the user has finished dictating, the user can enter the synchronous reproduction mode. In this mode, the speech of the user is played-back while at the same time the word, which was recognized for the segment of speech being played-back, is highlighted on the display. To this end, the speech of the user has been stored in a memory. Moreover, during the recognition, word boundaries are detected in the speech . For each word a begin mark, indicating the beginning of the word, and an end mark, indicating the end of the word, are stored. This enables an accurate synchronous reproduction of the speech and the highlighting on the display.
• the user can stop the synchronous reproduction and enter the editing and/or dictation mode.
• the user may enter the synchronous reproduction mode at any point in the text.
• a speech recognition system includes: a speech recognition system comprising: an input for receiving a speech representative signal; a first memory for storing a representation of the received signal suitable for audible reproduction; a speech recognizer operative to represent the received signal as a sequence of recognized words; a second memory for storing the sequence of recognized words, where each recognized word is stored in association with a marker indicating a correspondence between the word and a segment of the received signal in which the word was recognized; a controller operative to enable a user to review at least part of the sequence of recognized words by causing a synchronous reproduction of an audible and visible representation of the part of the sequence of recognized words, the synchronous reproduction including audibly reproducing a corresponding part of the received signal stored in the first memory and for each segment of the corresponding part of the received signal, at the moment when the segment is being audibly reproduced, indicating on a display a textual representation of a recognized word which corresponds to the segment; the correspondence being given by the markers stored in the second
• the editing instructions of the user may be received via any suitable form of input, including the keyboard (e.g. to insert/delete/replace a word or character(s) of a word), the mouse (e.g. to change formatting of a part of the text, like changing font, style or size, or to change an edit position), or via voice (e.g. to dictate one or more words/characters to insert/delete/replace a word or character(s) of a word or in the form of a voice command e.g. to change formatting of a part of the text or to change an edit position). It is no longer required that the user issues a dedicated instruction to leave the synchronous reproduction mode to be able to edit the text.
• the keyboard e.g. to insert/delete/replace a word or character(s) of a word
• the mouse e.g. to change formatting of a part of the text, like changing font, style or size, or to change an edit position
• voice e.g. to dictate one or more
• restart of the synchronous reproduction is automatic, implying that the user no longer needs to issues an explicit dedicated instruction to re-start the synchronous reproduction.
• the synchronous reproduction is automatically restarted after the user apparently has finished the editing in view of not having provided editing input for a certain period of time.
• the time-out is user-configurable providing the user a choice between the system quickly restarting the synchronous reproduction (with the risk that the user was still considering further editing operations) and the system restarting the synchronous reproduction more slowly (allowing the user more time to edit, at the expensive of an overall slower response).
• the user can still overrule the automatic behavior via an explicit instruction to stop a too quickly restarted reproduction or to restart a too slow restarting reproduction.
• the default time-out may be in the order of a few hundred milliseconds to a few seconds.
• the reproduction is restarted where it was paused. This will in normal situations allow for a smooth continuation of the reviewing.
• the reproduction is restarted at the last edited word. In most situations this position reflects the area of interest of the user, making it desired to restart the reproduction from that position.
• the user can simply change where in the sequence of recognized words the reproduction is active by indicating the desired position, e.g. by clicking the mouse at the desired position or via voice commands.
• the system is capable of dealing with those situations wherein the user wants to restart the dictation at the position currently reached in the reproduction mode. According to the invention, if the user simply starts his dictation by speaking (e.g. several seconds), the system no longer regards the voice input as being intended to edit (e.g. insert) a few words into the existing dictation, but instead exits the reproduction mode and goes into the dictation mode.
• a method of enabling reviewing a sequence of words recognized by a speech recognizer in a speech representative input signal includes the steps of: storing a representation of the received signal suitable for audible reproduction; using a speech recognizer to represent the received signal as a sequence of recognized words; storing the sequence of recognized words, where each recognized word is stored in association with a marker indicating a correspondence between the word and a segment of the received signal in which the word was recognized; enabling a user to review at least part of the sequence of recognized words by causing a synchronous reproduction of an audible and visible representation of the part of the sequence of recognized words, the synchronous reproduction including audibly reproducing a corresponding part of the received signal stored in the first memory and for each segment of the corresponding part of the received signal, at the moment when the segment is being audibly reproduced, indicating on a display a textual representation of a recognized word which corresponds to the segment; the correspondence being given by the markers stored in the second memory; detecting whether the user has provided an
• Figure 1 shows the elements of a typical speech recognizer
• Figure 2 illustrates HMM-based word models
• Figure 3 shows a block diagram of a speech recognition system according to the invention.
• Figure 4 illustrates an embodiment of using markers to indicate the correspondence between a word and a speech segment.
• Speech recognition systems such as large vocabulary continuous speech recognition systems, typically use a collection of recognition models to recognize an input pattern. For instance, an acoustic model and a vocabulary may be used to recognize words and a language model may be used to improve the basic recognition result.
• Figure 1 illustrates a typical structure of a large vocabulary continuous speech recognition system 100.
• the system 100 comprises a spectral analysis subsystem 110 and a unit matching subsystem 120.
• the speech input signal SIS
• OV observation vector
• the speech signal is digitized (e.g.
• an acoustic model provides the first term of equation (1).
• the acoustic model is used to estimate the probability P(Y
• a speech recognition unit is represented by a sequence of acoustic references. Various forms of speech recognition units may be used.
• a whole word or even a group of words may be represented by one speech recognition unit.
• a word model provides for each word of a given vocabulary a transcription in a sequence of acoustic references.
• a whole word is represented by a speech recognition unit, in which case a direct relationship exists between the word model and the speech recognition unit.
• a word model is given by a lexicon 134, describing the sequence of sub-word units relating to a word of the vocabulary, and the sub-word models 132, describing sequences of acoustic references of the involved speech recognition unit.
• a word model composer 136 composes the word model based on the subword model 132 and the lexicon 134.
• Figure 2A illustrates a word model 200 for a system based on whole-word speech recognition units, where the speech recognition unit of the shown word is modeled using a sequence of ten acoustic references (201 to 210).
• Figure 2B illustrates a word model 220 for a system based on sub-word units, where the shown word is modeled by a sequence of three sub-word models (250, 260 and 270), each with a sequence of four acoustic references
• HMMs Hidden Markov Models
• each recognition unit word model or subword model
• HMM state corresponds to an acoustic reference.
• Various techniques are known for modeling a reference, including discrete or continuous probability densities.
• Each sequence of acoustic references which relate to one specific utterance is also referred as an acoustic transcription of the utterance. It will be appreciated that if other recognition techniques than HMMs are used, details of the acoustic transcription will be different.
• a word level matching system 130 of Fig. 1 matches the observation vectors against all sequences of speech recognition units and provides the likelihoods of a match between the vector and a sequence. If sub-word units are used, constraints can be placed on the matching by using the lexicon 134 to limit the possible sequence of sub- word units to sequences in the lexicon 134. This reduces the outcome to possible sequences of words.
• a sentence level matching system 140 may be used which, based on a language model (LM), places further constraints on the matching so that the paths investigated are those corresponding to word sequences which are proper sequences as specified by the language model.
• the language model provides the second term P(W) of equation (1).
• P(W) the second term of equation (1).
• the language model used in pattern recognition may include syntactical and/or semantical constraints 142 of the language and the recognition task.
• a language model based on syntactical constraints is usually referred to as a grammar 144.
• N-gram word models are widely used.
• w 1 w 2 w 3 ...W j- ⁇ ) is approximated by P(w j
• bigrams or trigrams are used.
• W 1 W 2 W 3 ...WJ. 1 ) is approximated by P(WJ
• FIG. 3 shows a block diagram of the system according to the invention.
• the system includes a speech recognizer, such as the system 100 described above with reference to figures 1 and 2.
• a speech input signal (SIS) is fed into the speech recognizer.
• SIS speech input signal
• the speech is received in analog form via a microphone into a conventional audio card.
• the audio card samples the signal and converts it into a digital form.
• a digital microphone may be used, which feeds a digital speech signal directly into the system, for instance via USB.
• the digital representation of the speech is fed to the recognizer 100.
• a digital representation of the speech signal is also stored in a memory 330 for later reproduction during the synchronized playback/reproduction.
• the speech signal may be stored in any suitable form, such as a PCM sampled stream, or, preferably, a compressed form. Any compression technique suitable for compressing speech may be used.
• a dedicated memory 330 for storing speech may be used, including background storages.
• a part of the conventional solid state memory (like RAM) is used, which is also used by the speech recognizer.
• the stream of recognized words (recognized sentence RS) generated by the speech recognizer 100 is also stored in a memory 320. This memory forms preferably also part of the main memory.
• the recognizer determines the most likely word sequence and as part of this process determines the boundaries in the speech signal where those words begin and end. In itself, any suitable technique for determining the word boundaries may be used.
• a frequently used technique is based on backtracking to determine the most likely (sub-)word sequence. Such backtracking may be based on the well-known Viterbi algorithm.
• the system stores for each word also an indication (marker) of the beginning and/or the end of the word with respect to the stored speech signal. Preferably, those markers are stored in the same memory 320 as used for storing the words.
• Figure 4 shows that for each word wl, w2, w3, .. of the sequence of words 420 the beginning and end of a corresponding segment of the speech signal 410 is known.
• the correspondence is preferably maintained by storing for each word a reference to a beginning and an ending of the corresponding speech segment. For instance, word wl starts at Tl and ends at T2; w2 starts at T2 and ends at T3, etc.
• the markers can be seen as points in time since the speech signal is a time-sequential signal. This implies that the markers can be expressed as time-points in the signal 410, starting from a reference time point.
• the speech signal is stored as a sequence of samples, sampled at a fixed frequency, also a sample number can be used as a marker.
• Other s ⁇ itable markers may also be used, depending on the way of sampling, coding and storing of .he speech signal 410. It will be appreciated that since both the speech signal 410 and the word sequence 420 are time-sequential, it is in principle sufficient to store only one reference to a marker (either the begin or the end marker) for each word. For instance, assuming that playback starts at word wl and that only begin markers are stored, it is required to check the next word in sequence (w2) to determine the begin marker T2 of w2.
• highlighting of wl should be stopped and highlighting of T2 should be effected.
• a reference to the word of the sequence 420 which starts (or ends) at that position For instance, for marker Tl a reference (such as a pointer) is stored to word wl, for T2 a reference to w2, etc.
• the system also includes a controller 320 that controls the synchronous reproduction. Since normally the speech recognizer 100 is also implemented in software, it is preferred that the controller 320 also performs the functions of the speech recognizer 100. During the synchronous reproduction, the controller ensures that the highlighting on a display 340 and the audible reproduction is synchronous. Any suitable means may be used for the audible reproduction. If an audio card was used to receive the speech via a microphone from the user, preferably the same audio card is used for reproducing the speech via loudspeakers 350. Also digital speakers 350 may be used, for instance connected to the system via USB. According to the invention, the controller 310 is operative to cause a synchronous reproduction of an audible and visible representation of the sequence of recognized words.
• the user dictates a relatively large part of text, such as one or more paragraphs.
• the user reviews the recognition result.
• the dictation and recognition may have been 'real-time', e.g. by the user dictating directly into a word-processing or editing program, where a speech recognition program converts the speech into text.
• the reviewing user is normally th same as the user who performed the original dictation. It is also possible that the dictated speec h was recorded 'off-line' (e.g. into a high-quality memo-recorder) and was fed into the speech ecognizer at a later stage. In this case, the reviewing user may very well be another user, like a secretary. Normally the reviewing user can select at which position in the text the synchronous reproduction starts.
• This reproduction may involve all speech (and consequently all recognized words), but may also cover only a selected part of it.
• the involved part of the speech signal stored in memory 330 is played-back.
• the markers stored in memory 320 identify the sequence of segments which have been identified in the speech signal, where each segment relates to a recognized word. Whenever a next segment is being played-back, the textual representation of the word recognized for that segment is indicated on the display. The correspondence between the respective words and the respective speech segments is given by the markers stored in the memory 320. Normally, initially a window on the display is filled with the recognized text. In such a case, the recognized word corresponding to the speech segment can be indicated by highlighting the word on the screen.
• the high-lighting 'follows' the play-back of the speech.
• the available display area may not be sufficient to display a large area of recognized text. This may, for instance be the case, where the dictation is recognized on a mobile device, such as a PDA, advanced mobile phone, or mobile device for accessing Internet.
• the text may also 'slide' through a display area where the indicated text is the word at a specific position in the area, such as the center of the area.
• the speech is played-back at the same rate as it was recorded. If desired, also a slow-down or increase of the play-back rate may be selectable by the user. Techniques well-known from speech synthesis may be used to ensure that the quality of the speech reproduction does not degrade too much if the play-back rate differs from the recording rate.
• the controller 310 allows the reviewing user to correct the recognized text.
• the user may perform editing operations as normally provided by word processing systems.
• voice input for the editing is preferably possible. This may involve voice commands, e.g. to select text, change a format of text, etc. It may also involve dictating new words. A new word may be inserted in the text. It may also replace a selected text fragment.
• voice controlled editing and dictation are well known.
• the controller 310 is operative to detect whether the user has provided an editing instruction, while the synchronous reproduction is active. Normally, the operating system, under which the software of the controller runs, supports 'intercepting' the input from the user and detecting certain inputs and acting on it. If the controller detects such an editing instruction, it pauses the synchronous reproduction, and causes the editing instruction to be performed (e.g. by sending the instruction on to a word processing system).
• the editing instructions of the user may be received via any suitable form of input, including the keyboard, the mouse, or via voice.
• the controller is operative to detect when a user has finished editing.
• the controller uses a time-out mechanism to detect this.
• a timer is started at the moment of pausing the synchronous reproduction.
• the controller detects an editing operation (e.g., via the keyboard, mouse or voice)
• the timer is restarted as long as the timer has not expired yet.
• the controller ensures that the synchronous reproduction is restarted.
• the user can configure the time-out value. For instance, by specifying the time within a predetermined range such as 100 ms. to 2 seconds.
• the user may also be offered a choice from a limited number of options, such as quick, medium of slow restart.
• the controller stores the location, such as an indication of the word, which was reached during the reproduction at the moment of pausing. This enables the controller to restart the paused synchronous reproduction at that word or at the word immediately following.
• the controller detects which word or words has/have been edited during the pausing. It is generally known from editing systems how this can be detected. Having detected which words have been edited enables the controller to restart the reproduction immediately after the edited word(s). Since the user has already added new words or corrected wrongly recognized words, it is not preferred to start the synchronous reproduction with the already edited part of the text.
• the controller is operative to detect, during synchronous reproduction, that the user has indicated on the display a position different from the currently reproduced word. For instance, the user has positioned a mouse-controlled cursor in or before a different word. If the controller detects this, it causes the synchronous reproduction with a word corresponding to the position indicated by the user. It will be appreciated that if the user has located a cursor in between words (e.g. near a punctuation mark), the reproduction is preferably continued with the first word following that position.
• the controller also starts a timer at the moment the pause starts to determine how long the pause has lasted. If the pause lasts relatively long (e.g. several seconds), with more or less continuous editing input from the user, it is assumed that in fact the user is no longer correcting the recognized text, but is effectively entering new text. In that case it is no longer desired, that a pause in the input from the user results in restarting the synchronous reproduction. Instead, the synchronous reproduction mode is exited, and the dictation mode (optimized for entering new data) is started.
• the speech recognition system is implemented on a conventional computer, such as a PC or workstation.
• portable equipment such as PDAs, laptops or even telephones, are equipped with speech recognition.
• the functionality described by the invention is typically executed using the processor of the device, using conventional hardware, such as an audio card, for converting an analogue microphone signal into a digital data stream.
• conventional hardware such as an audio card
• DSPs may be used for performing the speech recognition.
• the processor such as a PC-type processor, micro-controller or DSP- like processor, can be loaded with a program to perform the steps according to the invention.
• This computer program product is usually loaded from a background storage, such as a harddisk or ROM.
• the computer program product can initially be stored in the background storage after having been distributed on a storage medium, like a CD-ROM, or via a network, like the public Internet.

Landscapes

• Engineering & Computer Science (AREA)
• Physics & Mathematics (AREA)
• Audiology, Speech & Language Pathology (AREA)
• Computational Linguistics (AREA)
• Health & Medical Sciences (AREA)
• Theoretical Computer Science (AREA)
• Artificial Intelligence (AREA)
• General Health & Medical Sciences (AREA)
• General Engineering & Computer Science (AREA)
• General Physics & Mathematics (AREA)
• Human Computer Interaction (AREA)
• Acoustics & Sound (AREA)
• Multimedia (AREA)
• Document Processing Apparatus (AREA)

Priority Applications (2)

Applications Claiming Priority (2)

Publications (1)

Family

ID=8241027

Family Applications (1)

Country Status (4)

Cited By (4)

Families Citing this family (40)

Citations (5)

Family Cites Families (18)

• 2000
  • 2000-12-07 EP EP10183069.3A patent/EP2261893B1/en not_active Expired - Lifetime
  • 2000-12-07 EP EP00981356.9A patent/EP1169678B1/en not_active Expired - Lifetime
  • 2000-12-07 WO PCT/EP2000/012447 patent/WO2001046853A1/en not_active Ceased
  • 2000-12-07 JP JP2001547300A patent/JP2003518266A/ja active Pending
  • 2000-12-19 US US09/741,671 patent/US6792409B2/en not_active Expired - Lifetime

Patent Citations (5)

Non-Patent Citations (2)

Also Published As

Similar Documents

Legal Events