EP1723636A1 - Determination de seuils de fiabilite et de rejet avec adaptation a l'utilisateur et au vocabulaire - Google Patents

Determination de seuils de fiabilite et de rejet avec adaptation a l'utilisateur et au vocabulaire

Info

Publication number
EP1723636A1
EP1723636A1 EP05707860A EP05707860A EP1723636A1 EP 1723636 A1 EP1723636 A1 EP 1723636A1 EP 05707860 A EP05707860 A EP 05707860A EP 05707860 A EP05707860 A EP 05707860A EP 1723636 A1 EP1723636 A1 EP 1723636A1
Authority
EP
European Patent Office
Prior art keywords
speaker
recognition
confidence
recognized
user
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Withdrawn
Application number
EP05707860A
Other languages
German (de)
English (en)
Inventor
Tobias Stranart
Andreas Schröer
Michael Wandinger
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Siemens AG
Original Assignee
Siemens AG
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Siemens AG filed Critical Siemens AG
Publication of EP1723636A1 publication Critical patent/EP1723636A1/fr
Withdrawn legal-status Critical Current

Links

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
    • G10L15/06—Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice
    • G10L15/065—Adaptation
    • G10L15/07—Adaptation to the speaker
    • 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
    • G10L15/06—Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice

Definitions

  • the underlying technical procedure for speaker-independent speech recognition consists in the classification of recognition results in categories for the security of the recognition, e.g. [recognized with certainty, recognized with uncertainty, not in vocabulary].
  • categories for the security of the recognition e.g. [recognized with certainty, recognized with uncertainty, not in vocabulary].
  • the number and naming of these categories vary depending on the speech recognition technology used, as well as their treatment in the speech application. So it is e.g. conceivable that a voice application for uncertainly recognized words makes a query to the user.
  • a speech recognition system there is therefore the problem of providing as precise and error-free a division as possible into one of the above-mentioned categories for each recognition result.
  • the basis of the division into categories when classifying recognition results is a so-called confidence measure that is calculated by the speech recognition system for each recognition result.
  • the literature offers a variety of algorithms for calculating this figure.
  • Relevant for this invention is the framework in which suitable confidence measure threshold values are determined. These define the categories mentioned above for the security of the detection. It should be noted that a well chosen one In addition to the language and the modeling used (eg Hidden Markov Model), the threshold also depends on the speaker and the recognition vocabulary.
  • the speaker-independent detection is based e.g. on Hidden Markov modeling. It offers comfort for the user, since no special training (audition, enrollment) of the words to be recognized is required.
  • the vocabulary to be recognized must be known a priori.
  • phonetic or graphemic information about the words to be recognized There are standard methods for the graphemics of a word, i.e. convert its written form into its phonetic form, which is the form required by the speech recognition system.
  • speaker-dependent recognition is the dialing of a cell phone by name.
  • the names from the phone book were typically trained beforehand depending on the speaker (SD enrollment) - an acoustic model for recognition is generated based on the spoken form of a word.
  • the standard methods of speaker-independent detection do not work here, the thresholds of si detection are not transferable.
  • Predefined confidence measure threshold values for speaker-dependent vocabularies are typically not adapted to a speaker or a vocabulary and are therefore suboptimal per se. It can even go so far that they cannot be used at all.
  • Known approaches include the - less desirable - direct influence of the user on the thresholds, ie he is forced to influence the % sharpness x of the rejection of the detection system itself.
  • speaker-independent modeling of a word or vocabulary is adapted to a speaker through adaptive training.
  • the goal is to improve the recognition rate by recording speaker-specific characteristics.
  • the adaptation to a speaker can take place at the phoneme level or at the word level. Similar to the SD case, no solutions are known to take into account the impact of the additional training / adaptation process on the confidence threshold.
  • the object of the invention is to enable a meaningful determination of confidence measures and confidence thresholds, particularly in the case of speaker-dependent and speaker-adaptive speech recognition.
  • a target recognition result is provided in a method for determining confidence measures in speech recognition for a recognition process.
  • the recognition process is carried out and the confidence measure of the target recognition result is determined on the basis of information which is obtained when the recognition process is carried out.
  • a confidence threshold is preferably defined taking into account the confidence measure. This procedure is used in particular for speaker-dependent or speaker-adaptive speech recognition.
  • a confidence threshold can already be given and the confidence threshold is adjusted taking into account the confidence measure.
  • the confidence threshold is advantageously a confidence threshold for classifying recognition results into categories.
  • the categories include, for example, a category in which an utterance to be recognized is recognized as certain, a category in which an utterance to be recognized is recognized as unsure, and / or a category to which utterances are assigned that are not to be recognized cognitive vocabulary.
  • the procedure is a procedure for speaker-independent or speaker-adaptive speech recognition
  • the target is
  • Predetermined recognition result because in the recognition process a statement given to the user is recognized.
  • the method is a method for speaker-dependent speech recognition
  • the user is advantageously asked to speak an utterance to be recognized at least twice, the recognition result being the one time Provides target recognition result, while at other times, among other things, the confidence measure of the target recognition result is determined.
  • An arrangement that is set up to carry out one of the described methods can be implemented, for example, by programming and setting up a data processing system using means that belong to the aforementioned method steps.
  • a program product for a data processing system which contains code sections with which one of the described methods can be carried out on the data processing system, can be implemented by suitably implementing the method in a programming language and translating it into
  • a program product is understood to mean the program as a tradable product. It can be in any form, for example on paper, a computer-readable data medium or distributed over a network.
  • FIG. 1 shows a flow diagram of a speaker-adaptive training process with adjustment of confidence thresholds
  • FIG. 2 shows a flowchart of a speaker-dependent training process with adjustment of confidence thresholds
  • Figure 3 shows a detection process with adjustment of confidence thresholds.
  • the method presented here is based on using information that is obtained when a recognition process is carried out in which the target recognition result is known.
  • speaker adaptive training the user is shown the word to be spoken, so the target is known. This also applies to speaker-specific training, since the user speaks the word that is to be added to the vocabulary of speaker-dependent speech recognition. In this case of speaker-dependent speech recognition, however, it is necessary that the word to be added is spoken twice, since otherwise there is no word model that could serve as a target recognition result. However, this condition is met in very many cases, since most current speech recognition systems require a double training of speaker-dependent speech recognition for a variety of reasons.
  • the method can generally also be used in a recognition process if there is knowledge as to whether the recognition result is correct or whether this knowledge can be derived, for example from the user's reaction. This applies to all of the above cases of speaker-independent, speaker-dependent and speaker-adaptive speech recognition.
  • the utterances used for determining or adapting / improving the identity threshold are specific to the speaker and the vocabulary to be recognized. This is precisely the weak point of the methods described in the prior art.
  • FIG. 1 shows a speaker-adaptive training process with adjustment of the confidence thresholds.
  • the user carries out an adaptation process by speaking the words specified by the speech recognition system.
  • a recognition process is carried out for each utterance of the user and the confidence measure of the target recognition result is determined.
  • the confidence thresholds for categorizing are either defined or, if an iteration value already exists, adjusted accordingly.
  • these confidence thresholds are optimally adapted to the user.
  • they are also specific to the vocabulary of recognition and thus enable, for example, an improved rejection.
  • FIG. 2 shows a speaker-dependent training process with adjustment of the confidence thresholds.
  • the user first speaks the word to be added to the vocabulary.
  • the recognition system uses this data to generate a speaker-dependent reference word model, which is provisionally adopted in the vocabulary. Then the word is spoken again by the user.
  • This second pass is required by most of the speech recognition systems on the market anyway for reasons of securing, verifying and increasing the recognition performance. With this second utterance, a recognition process is carried out, and the confidence measure for the word model of the first pass is determined, which represents the target recognition result.
  • the confidence thresholds for the category classification are either defined or, if an iteration value already exists, adjusted accordingly.
  • FIG. 3 shows the sequence in a recognition process with adjustment of the confidence thresholds.
  • the user performs a recognition process.
  • the target result is known or can be derived the.
  • the confidence measure for the utterance is determined for this target recognition result, and the confidence thresholds for categorizing are adjusted accordingly.
  • these confidence thresholds are optimally adapted to users and vocabulary.
  • a special feature is the use of recognition processes with a known target recognition result.
  • the method allows their evaluation to determine a specific measure of identity. This means that for the first time a realistic result classification is possible for speaker-dependent and speaker-adaptive speech recognition.
  • the method can be used with different algorithms described in the literature for calculating a confidence measure.

Landscapes

  • Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Human Computer Interaction (AREA)
  • Physics & Mathematics (AREA)
  • Acoustics & Sound (AREA)
  • Multimedia (AREA)
  • Machine Translation (AREA)

Abstract

Lors d'un processus de reconnaissance vocale, un résultat de reconnaissance théorique peut être déduit ou est déjà fourni. Le processus de reconnnaissance est effectué, et une mesure de fiabilité du résultat de reconnaissance théorique est déterminée.
EP05707860A 2004-03-12 2005-01-27 Determination de seuils de fiabilite et de rejet avec adaptation a l'utilisateur et au vocabulaire Withdrawn EP1723636A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
DE102004012206 2004-03-12
PCT/EP2005/050342 WO2005088607A1 (fr) 2004-03-12 2005-01-27 Determination de seuils de fiabilite et de rejet avec adaptation a l'utilisateur et au vocabulaire

Publications (1)

Publication Number Publication Date
EP1723636A1 true EP1723636A1 (fr) 2006-11-22

Family

ID=34960275

Family Applications (1)

Application Number Title Priority Date Filing Date
EP05707860A Withdrawn EP1723636A1 (fr) 2004-03-12 2005-01-27 Determination de seuils de fiabilite et de rejet avec adaptation a l'utilisateur et au vocabulaire

Country Status (3)

Country Link
US (1) US8874438B2 (fr)
EP (1) EP1723636A1 (fr)
WO (1) WO2005088607A1 (fr)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110457630A (zh) * 2019-07-30 2019-11-15 北京航空航天大学 一种开源社区异常点赞用户的识别方法及系统

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070124147A1 (en) * 2005-11-30 2007-05-31 International Business Machines Corporation Methods and apparatus for use in speech recognition systems for identifying unknown words and for adding previously unknown words to vocabularies and grammars of speech recognition systems
US8010358B2 (en) * 2006-02-21 2011-08-30 Sony Computer Entertainment Inc. Voice recognition with parallel gender and age normalization
US7778831B2 (en) * 2006-02-21 2010-08-17 Sony Computer Entertainment Inc. Voice recognition with dynamic filter bank adjustment based on speaker categorization determined from runtime pitch
US8239203B2 (en) * 2008-04-15 2012-08-07 Nuance Communications, Inc. Adaptive confidence thresholds for speech recognition
US8639508B2 (en) * 2011-02-14 2014-01-28 General Motors Llc User-specific confidence thresholds for speech recognition
EP3065132A1 (fr) * 2015-03-06 2016-09-07 ZETES Industries S.A. Méthode et système de détermination de validité d'un élément d'un résultat de reconnaissance vocale

Family Cites Families (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4181821A (en) 1978-10-31 1980-01-01 Bell Telephone Laboratories, Incorporated Multiple template speech recognition system
US5625748A (en) * 1994-04-18 1997-04-29 Bbn Corporation Topic discriminator using posterior probability or confidence scores
US5737489A (en) 1995-09-15 1998-04-07 Lucent Technologies Inc. Discriminative utterance verification for connected digits recognition
US6012027A (en) 1997-05-27 2000-01-04 Ameritech Corporation Criteria for usable repetitions of an utterance during speech reference enrollment
DE69829187T2 (de) 1998-12-17 2005-12-29 Sony International (Europe) Gmbh Halbüberwachte Sprecheradaptation
CA2748396A1 (fr) * 1999-10-19 2001-04-26 Sony Electronics Inc. Systeme de commande d'interface en langage naturel
US6473735B1 (en) * 1999-10-21 2002-10-29 Sony Corporation System and method for speech verification using a confidence measure
US6778959B1 (en) * 1999-10-21 2004-08-17 Sony Corporation System and method for speech verification using out-of-vocabulary models
US7941313B2 (en) * 2001-05-17 2011-05-10 Qualcomm Incorporated System and method for transmitting speech activity information ahead of speech features in a distributed voice recognition system
US7203643B2 (en) * 2001-06-14 2007-04-10 Qualcomm Incorporated Method and apparatus for transmitting speech activity in distributed voice recognition systems
US7089188B2 (en) * 2002-03-27 2006-08-08 Hewlett-Packard Development Company, L.P. Method to expand inputs for word or document searching
US6983246B2 (en) * 2002-05-21 2006-01-03 Thinkengine Networks, Inc. Dynamic time warping using frequency distributed distance measures
EP1377000B1 (fr) * 2002-06-11 2009-04-22 Swisscom (Schweiz) AG Procédé mis en oeuvre dans un système annuaire automatique actionné par la parole
US7363224B2 (en) * 2003-12-30 2008-04-22 Microsoft Corporation Method for entering text

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See references of WO2005088607A1 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110457630A (zh) * 2019-07-30 2019-11-15 北京航空航天大学 一种开源社区异常点赞用户的识别方法及系统
CN110457630B (zh) * 2019-07-30 2022-03-29 北京航空航天大学 一种开源社区异常点赞用户的识别方法及系统

Also Published As

Publication number Publication date
WO2005088607A1 (fr) 2005-09-22
US20070213978A1 (en) 2007-09-13
US8874438B2 (en) 2014-10-28

Similar Documents

Publication Publication Date Title
DE69831114T2 (de) Integration mehrfacher Modelle für die Spracherkennung in verschiedenen Umgebungen
DE602006000090T2 (de) Konfidenzmaß für ein Sprachdialogsystem
DE60302407T2 (de) Umgebungs- und sprecheradaptierte Spracherkennung
DE60125542T2 (de) System und verfahren zur spracherkennung mit einer vielzahl von spracherkennungsvorrichtungen
DE69722980T2 (de) Aufzeichnung von Sprachdaten mit Segmenten von akustisch verschiedenen Umgebungen
DE112010005959B4 (de) Verfahren und System zur automatischen Erkennung eines Endpunkts einer Tonaufnahme
DE69818231T2 (de) Verfahren zum diskriminativen training von spracherkennungsmodellen
DE69635655T2 (de) Sprecherangepasste Spracherkennung
DE112018002857T5 (de) Sprecheridentifikation mit ultrakurzen Sprachsegmenten für Fern- und Nahfeld-Sprachunterstützungsanwendungen
DE60004331T2 (de) Sprecher-erkennung
DE60128270T2 (de) Verfahren und System zur Erzeugung von Sprechererkennungsdaten, und Verfahren und System zur Sprechererkennung
DE69924596T2 (de) Auswahl akustischer Modelle mittels Sprecherverifizierung
EP1723636A1 (fr) Determination de seuils de fiabilite et de rejet avec adaptation a l'utilisateur et au vocabulaire
EP1611568A1 (fr) Reconnaissance de mots isoles en trois etapes
EP1264301B1 (fr) Procede pour reconnaitre les enonces verbaux de locuteurs non natifs dans un systeme de traitement de la parole
DE10006930B4 (de) System und Verfahren zur Spracherkennung
DE60034772T2 (de) Zurückweisungsverfahren in der spracherkennung
DE10119284A1 (de) Verfahren und System zum Training von jeweils genau einer Realisierungsvariante eines Inventarmusters zugeordneten Parametern eines Mustererkennungssystems
EP1199704A2 (fr) Sélection d'une séquence alternative de mots pour une adaptation discriminante
DE10304460B3 (de) Generieren und Löschen von Aussprachevarianten zur Verringerung der Wortfehlerrate in der Spracherkennung
DE102010040553A1 (de) Spracherkennungsverfahren
WO2001086634A1 (fr) Procede pour produire une banque de donnees vocales pour un lexique cible pour l'apprentissage d'un systeme de reconnaissance vocale
DE60014583T2 (de) Verfahren und vorrichtung zur integritätsprüfung von benutzeroberflächen sprachgesteuerter geräte
DE102004017486A1 (de) Verfahren zur Geräuschreduktion bei einem Sprach-Eingangssignal
WO2005069278A1 (fr) Procede et dispositif pour traiter un signal vocal pour la reconnaissance vocale robuste

Legal Events

Date Code Title Description
PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

Free format text: ORIGINAL CODE: 0009012

17P Request for examination filed

Effective date: 20060731

AK Designated contracting states

Kind code of ref document: A1

Designated state(s): DE FR GB IT

DAX Request for extension of the european patent (deleted)
RBV Designated contracting states (corrected)

Designated state(s): DE FR GB IT

17Q First examination report despatched

Effective date: 20070625

RAP1 Party data changed (applicant data changed or rights of an application transferred)

Owner name: SIEMENS AKTIENGESELLSCHAFT

RAP1 Party data changed (applicant data changed or rights of an application transferred)

Owner name: SIEMENS AKTIENGESELLSCHAFT

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: THE APPLICATION IS DEEMED TO BE WITHDRAWN

18D Application deemed to be withdrawn

Effective date: 20140801