SOURCES SOUGHT
U -- Creation of Pronuciation Training Courseware for the non-native speaker.
- Notice Date
- 8/23/2010
- Notice Type
- Sources Sought
- NAICS
- 611710
— Educational Support Services
- Contracting Office
- USAR Contracting Center - West (POM), Bldg 4385, Suite 2041, 400 Gigling Road, Seaside, CA 93955
- ZIP Code
- 93955
- Solicitation Number
- W9124N-10-SS-0001
- Response Due
- 9/20/2010
- Archive Date
- 11/19/2010
- Point of Contact
- Todd Bales, 831-242-4078
- E-Mail Address
-
USAR Contracting Center - West (POM)
(todd.s.bales@us.army.mil)
- Small Business Set-Aside
- N/A
- Description
- Automatic Speech Recognition (ASR) Engine and Feedback System to be integrated to DLIFLCs Computer-Assisted Pronunciation Training (CAPT) System for Native Speakers of English Learning Chinese Mandarin, Dari, French, Iraqi, Korean, Pashto, Persian-Farsi, German, Portuguese (European), Russian, Spanish, Urdu, or Uzbek FOR INFORMATIONAL PURPOSES ONLY THIS IS NOT A SOLICITATION ANNOUNCEMENT. This sources sought solicitation is for information and planning purposes only and is issued for the purpose of market research in accordance with FAR Part 10. The Dept of the Army, Contracting Center West Monterey, CA is currently requesting information to determine the existence of viable large and small businesses that are interested and capable of performing the work described herein. Responses are welcome from all business sources. This sources sought solicitation is not limited to any socioeconomic program and/or business set-side. As a result of this notice, the Government may issue a Request for Proposal; however there is no solicitation available at this time. This notice is for planning purposes only, and does not constitute an Invitation for Bids, a Request for Proposals, Solicitation, Request for Quotes, or an indication that the Government will contract for the items contained herein. This notice is not to be construed as a commitment on the part of the Government to award a contract, nor does the Government intend to pay for any information submitted as a result of this notice. The Government will not reimburse respondents for any cost associated with submission of the information being requested or reimburse expenses incurred to interested parties for responses to this sources sought notice. Any responses received will not be used as a proposal. This is not a request for proposals and in no way obligates the Government to award any contract. OBJECTIVES The Defense Language Institute Foreign Language Center (DLIFLC) is conducting market research to assess the capabilities of building a pronunciation training courseware for the non-native speaker (NNS) whose primary language of use is English, and who is learning one of the following languages as a foreign language: Chinese Mandarin, Dari, French, Iraqi, Korean, Pashto, Persian-Farsi, German, Portuguese (European), Russian, Spanish, Urdu, and Uzbek. We are inviting the proposals for the computer assisted pronunciation training program equipped with automatic speech recognition engine and the feedback system. DLIFLC is seeking innovative ways to provide meaningful and effective computer assisted pronunciation practice to military linguists and civilian government employees who need foreign language training but cannot attend face-to-face classroom lessons. The newly developed CAPT will be integrated into Headstart2, which is a distributed learning program developed by government (See the next section for more information). The developed system should be able to provide meaningful and effective pronunciation training to native English speakers learning one of the target languages listed above. In doing so, DLIFLC is seeking innovative ways to provide intelligent and effective pronunciation training and feedback to the learners. The desired outcome of the intelligent CAPT and feedback system is that learners can self-correct the target language pronunciation and approximate their pronunciation to the native-like sound. In recent years, commercial language learning software providers have offered visualization courseware which assists students to compare their own recorded utterances to wave forms displayed on a screen. Its purpose was that students approximate a model sound wave and analyze the deviation between the native speaker norm and their own recorded utterances. However, it was found that language learners did not always clearly understand how to interpret sound wave representation. Although more developed programs is available in the market using animated representations of a mouth or a talking head to illustrate the production of speech sounds, it is still depending on how learners interpret the visual feedback on articulation organs. To overcome the limitations of the currently available CAPT and feedback system, DLIFLC is inviting information on state-of-art speech recognition engine for NNS. The developed system should be able to isolate pronunciation errors of NNS in phone level as well as provide a customized feedback based on what is actually spoken by learners. HEADSTART 2 As one of the excellences of distributed learning programs produced and managed by DLIFLC, Headstart2 is an introductory level self-study language training designed for those who may have neither any formal instruction nor any previous experience on the target language. Its main purpose is to familiarize deploying military personnel with the target language and the target culture using real-life military scenarios within limited training time. Headstart2 incorporates 750 of the most commonly needed phrases by troops on the ground. This computer-based program uses human-to-avatar interaction, games, word scrambles, and other interactive exercises to draw the user into the learning program. In this 80-100hour course, troops are introduced to the target language writing system in the first two tasks. Then, it gradually guides the learner to read street signs, understand basic greetings, and ask and respond to simple questions. Headstart2 can be downloaded onto devices such as iPods, laptop computers or desktop PCs. Currently, DLIFLC is working on general access of Headstart2. Headstart2 for Chinese Mandarin, Dari, French, Iraqi, Korean, Pashto, Persian-Farsi, German, Portuguese (European), Russian, Spanish, Urdu, and Uzbek are either currently available or in production. DLIFLC has in-house cadre of experts who are capable of developing full scale instruction design (i.e. curriculum, lesson, activity, etc.) and web interface. Language and curriculum experts are experienced with designing the instruction for military specific purpose. However, the current Headstart2 focuses on listening and reading comprehension and lacks the speaking training module. TECHNICAL REQUIREMENTS DLIFLC expects the offered system to satisfy the following specific requirements. In terms of system architecture, each component should be encapsulated in structure so that upgrades can be easily made for future if necessary. Recognition: Automatic Speech Recognition (ASR) Tasks in CAPT module are likely to be: reading off the screen, repeating, multiple choice questions, etc. Given that the types of tasks in Headstart 2 is highly constrained, it is not expected the ASR system to handle speech input high in linguistic entropy. However, the challenging aspect of developing ASR lies on the proficiency level of the expedited Headstart2 users, who are likely to be at ILR Level 0 or 0+ and are trying to reach more or less Level 1 after the training completed. It is a well-known problem that state-of-art speech recognition system encounter drastic increase in word error rate (WER), when they are faced with non-native speech. One of the key requirements of ASR system for Headstart2 will be its robustness in handling very-low-proficiency NNS speech in one of the target languages. In the information submitted for the sources sought, offeror shall address a specific methodology how very low proficiency NNSs speech will be handled in the system. Such as, offer may take HMM trained with NNS speech or standard speaker adaptation techniques like MLLR, MMI, etc. If the proposed methodology requires NNS speech sample, please note that DLIFLC may or may not assist offeror in collecting the NNS speech samples. Offer has to have his/her own plan to collect the sufficient amount of data required. Localization: Localization module should be able to determine which parts of a users response contained errors or poorly pronounced words or segment. It may be accomplished by querying each phone model as to how well it fit the region of speech identified as containing that phone. In order for accurate isolation of the learner error, the standard native speaker model of the target languages must demonstrate its robust performance yet flexible to those individual speakers variations, such as gender, age, or region. For validation, DLIFLC will prepare native speech sample in advance using a portion of the Headstart2 vocabulary and phone error rate (PER) will be measured. The information submitted by offeror may include proven or estimated PER of the offerors ASR. The quality of the native speech model is critical in CAPT otherwise the learners will be confused and the training will have negative effect to the learners. In setting the acceptance thresholds, it is preferred that the learners have a choice of setting a threshold depending on their capability. The choices may be given as a form of a category (e.g. native-like, intelligent or minimally intelligent) or a form of percentage (e.g. 80%). Lexical Capacity To meet the requirements of Headstart2, a minimally suitable ASR product should be able to handle a preselected base vocabulary of about 1,000 words. Diagnosis and Feedback Innovative CAPT courseware should be equipped with a system determining how the student mispronounced a word or phoneme. Most of the errors will be considered to stem from: i)Native/First Language (L1) influence ii)Novel features in Foreign/Second Language (L2) iii)Orthography iv)Polysemous lexicon v)Other inexplicable causes For the best result, each of these kinds of diagnoses may require different forms of feedback in order to best help the users. Feedback bank should base on the contrastive linguistic research and the data analysis of NNS speech done by experienced researchers in the field of linguistics. Depending on the target language, different types of errors are expected (e.g. tone error, re-syllabification, addition or deletion of a phone, vowel duration, allophones, etc.) To maximize the benefits of CAPT, learners should be able to understand the feedback provided without requiring previous knowledge on theoretical phonology or phonetics. To facilitate learners understanding, feedback may incorporate animated articulatory organs like mid-sagittal sections of the human head for tongue and velum positions. DLIFLC is also inviting all other innovative ways that will assist the self-correction process of NNS pronunciation. INTSTRUCTIONAL REQUIREMENTS In addition to the ASR and feedback system, offerors may be asked to provide full-scale CAPT lesson -including instruction, activity, and assessment (depending on government cost-effectiveness or availability of in-house development resource at time.) If the requirement is to build a full scale CAPT lesson, offerors shall consider the followings in designing the courseware. Activities Activities should engage the learner in interactive communication to the largest possible extent. Typically, these activities teach the discrimination of minimal pairs or consist of imitating native speaker models. Question and answer dialogs provide practicing exercises in real world contexts. Quality of Input The tutoring software will present a variety of native speaker models. Students should be exposed to high quality input from multiple native speakers (gender/age variations) to represent a range of speech sounds. For those target languages that are highly similar to other languages in neighboring regions (e.g. dialect), the authenticity of the target language should be emphasized, e.g. Modern Standard Arabic vs. Iraqi. Assessment of Student Performance Students should be able to produce speech that is intelligible to most native speakers of the foreign language. CAPT courseware will evaluate learner performance and record their progress as the training proceeds. Adequate levels of both positive and negative feedback should be provided wherever appropriate. Feedback should be easy to understand and constructive. Students should be able to compare their own voice to that of a native speaker and receive explicit guidance in improving their speech performance. Both feedback and scoring should be easily understood without prior theoretical Learner Autonomy Learners should be able to work independently and easily control the recordings. Navigation of CAPT should be intuitive and self-explanatory. In addition, students should be able to exit the instruction and resume at the point they ended. They should also be able to go directly to any new subject they intend to practice. CAPT should also be able to provide a snapshot of where they are in the overall training process and the progress they have made to the point. CALL FOR INFORMATION Any interested vendors should submit a Capability Statement demonstrating the professional expertise and technology to achieve the specified requirements addressed above. The statement may include the following information but is not limited to: "Methodologies or strategies to achieve the specified objectives and requirements addressed above. "Previous development work of ASR system (specially, for NNS) "Previous experience with similar projects. "Patented technology (including technology that is patent pending), if any. "Relevant background, qualifications, and industry-specific awards. This solicitation for product offerings is for INFORMATIONAL PURPOSES ONLY and does not constitute an implied contract with any vendor. Developing an Computer Assisted Pronunciation Training (CAPT) and Feedback System requires a wide range of expertise, including foreign language teaching, phonology and phonetics, second language acquisition, Computational Linguistics, Natural Language Processing (NLP), Speech Processing, Information Technology (IT), etc. Not all interested vendors may currently have all the required expertise. If not, they may include strategies that outline how they propose to meet those requirements. If more than one vendor works in partnership with another vendor, the statement must indicate which party is responsible for providing which specific types of expertise. In addition, if involved in a partnership, the statement must indicate any previous collaborative experience among or between the vendors. Send Capability Statement as outlined above to: Todd S. Bales Contract Specialist 400 Gigling Road, Suite 2202 Seaside CA 93955 or e-mail todd.s.bales@us.army.mil (Place 'DLI-Speech Recognition' in header)
- Web Link
-
FBO.gov Permalink
(https://www.fbo.gov/notices/5c0cf1931483f90bff2478128de387d6)
- Place of Performance
- Address: USAR Contracting Center - West (POM) Bldg 4385, Suite 2041, 400 Gigling Road Seaside CA
- Zip Code: 93955
- Zip Code: 93955
- Record
- SN02250381-W 20100825/100823234557-5c0cf1931483f90bff2478128de387d6 (fbodaily.com)
- Source
-
FedBizOpps Link to This Notice
(may not be valid after Archive Date)
| FSG Index | This Issue's Index | Today's FBO Daily Index Page |