In this                 newsletter:

-              Mixer 6 now available!  -
         
          -         
           Fall 2013 LDC Data           Scholarship Program - deadline approaching!  -

-           LDC at Interspeech 2013,           Lyon France  -

New publications:

-          GALE Phase 2 Chinese         Broadcast Conversation Parallel Text Part 2   -
     
      -  MADCAT Phase 3 Training Set  -

-  Mixer 6 Speech  -

      The release of Mixer 6 Speech this month marks the first time in       close to a decade that LDC has made available a large-scale speech       training data collection. Representing more than 15,000 hours of       speech from over 500 speakers, Mixer 6 follows in the footsteps of       the Switchboard and Fisher studies by providing a large database       of rich telephone conversations with the addition of subject       interviews and transcript readings.  Participants were native       American English speakers local to the Philadelphia area,       providing further scope for a variety of research tasks.  Mixer 6       Speech is a members-only release and a great reason to join the       consortium. In addition to this substantial resource, members        enjoy rights to other data released in 2013 and can license older       publications at reduced fees. Please see the full
        description of Mixer 6 Speech.

Fall         2013 LDC Data Scholarship Program - deadline approaching!

The deadline for the Fall 2013 LDC Data       Scholarship Program is one month away!   Student applications are       being accepted now through September
        16, 2013, 11:59PM EST.  The LDC Data       Scholarship program provides university students with access to       LDC data at no cost.  This program is open to students pursuing       both undergraduate and graduate studies in an accredited college       or university. LDC Data Scholarships are not restricted to any       particular field of study; however, students must demonstrate a       well-developed research agenda and a bona fide inability to pay.       
     
      Students will need to complete an application which consists of a       data use proposal and letter of support from their adviser.  For       further information on application materials and program rules,       please visit the LDC Data         Scholarship page. 

Students can email their applications to the LDC Data         Scholarship program. Decisions will be sent by email from       the same address.

      LDC at           Interspeech 2013, Lyon France

LDC will once         again be exhibiting at Interspeech held this year August 25-29 in Lyon.         Please stop by LDC’s booth to to learn about recent developments         at the Consortium, including new publications.

Also, be on         the lookout for the following presentations:

·                 Speech Activity Detection on YouTube Using Deep Neural Networks      

     

o   Neville           Ryant, Mark Liberman, Jiahong Yuan (all LDC)

     

o   Monday           26 August, Poster 6,  16.00
          – 18.00

     

o   Room:           Forum 6
       

   

·         The         Spectral Dynamics of Vowels in Mandarin Chinese

     

o   Jiahong           Yuan

     

o   Tuesday           27 August, Oral 17, 14.00 – 16.00

     

o   Room:           Gratte-Ciel 3
         
       

   

·                 Automatic Phonetic Segmentation using Boundary Models

     

o   Jiahong           Yuan (LDC), Neville Ryant (LDC), Mark Liberman (LDC), Andrea           Stolcke, Vikramjit Mitre, Wen Wang

     

o             Wednesday 28 August, Oral 32, 14.00 – 16.00

     

o   Room:           Gratte-Ciel 3

   

LDC will         continue to post conference updates via our Facebook page. We hope to see you there!

New
          publications:

(1)GALE Phase
        2 Chinese Broadcast Conversation Parallel Text Part 2 was       developed by LDC. Along with other corpora, the parallel text in       this release comprised training data for Phase 2 of the DARPA GALE       (Global Autonomous Language Exploitation) Program. This corpus       contains Chinese source text and corresponding English       translations selected from broadcast conversation (BC) data       collected by LDC in 2005-2007 and transcribed by LDC or under its       direction.

This release includes 20 source-translation       document pairs, comprising 152,894 characters of Chinese source       text and its English translation. Data is drawn from six distinct       Chinese programs broadcast in 2005-2007 from Phoenix TV, a Hong       Kong-based satellite television station. Broadcast conversation       programming is generally more interactive than traditional news       broadcasts and includes talk shows, interviews, call-in programs       and roundtable discussions. The programs in this release focus on       current events topics.

The data was transcribed by LDC staff and/or       transcription vendors under contract to LDC in accordance with       Quick Rich Transcription guidelines developed by LDC. Transcribers       indicated sentence boundaries in addition to transcribing the       text. Data was manually selected for translation according to       several criteria, including linguistic features, transcription       features and topic features. The transcribed and segmented files       were then reformatted into a human-readable translation format and       assigned to translation vendors. Translators followed LDC's       Chinese to English translation guidelines. Bilingual LDC staff       performed quality control procedures on the completed       translations.

GALE Phase 2 Chinese Broadcast Conversation       Parallel Text Part 2 is distributed via web download.

2013 Subscription Members will automatically       receive two copies of this data on disc. 2013 Standard Members may       request a copy as part of their 16 free membership corpora.        Non-members may license this data for US$1750.

*

(2)MADCAT (Multilingual
        Automatic Document Classification Analysis and Translation)         Phase 3 Training Set contains all training data created by       LDC to support Phase 3 of the DARPA MADCAT Program. The data in       this release consists of handwritten Arabic documents, scanned at       high resolution and annotated for the physical coordinates of each       line and token. Digital transcripts and English translations of       each document are also provided, with the various content and       annotation layers integrated in a single MADCAT XML output.

The goal of the MADCAT program is to       automatically convert foreign text images into English       transcripts. MADCAT Phase 3 data was collected from Arabic source       documents in three genres: newswire, weblog and newsgroup text.       Arabic speaking scribes copied documents by hand, following       specific instructions on writing style (fast, normal, careful),       writing implement (pen, pencil) and paper (lined, unlined). Prior       to assignment, source documents were processed to optimize their       appearance for the handwriting task, which resulted in some       original source documents being broken into multiple pages for       handwriting. Each resulting handwritten page was assigned to up to       five independent scribes, using different writing conditions.

The handwritten, transcribed documents were       next checked for quality and completeness, then each page was       scanned at a high resolution (600 dpi, greyscale) to create a       digital version of the handwritten document. The scanned images       were then annotated to indicate the physical coordinates of each       line and token. Explicit reading order was also labeled, along       with any errors produced by the scribes when copying the text.

The final step was to produce a unified data       format that takes multiple data streams and generates a single       MADCAT XML output file which contains all required information.       The resulting madcat.xml file contains distinct components: a text       layer that consists of the source text, tokenization and sentence       segmentation; an image layer that consists of bounding boxes; a       scribe demographic layer that consists of scribe ID and partition       (train/test); and a document metadata layer.

This release includes 4,540 annotation files in       both GEDI XML and MADCAT XML formats (gedi.xml and madcat.xml)       along with their corresponding scanned image files in TIFF format.      
     
      MADCAT Phase 3 Training Set is distributed on one DVD-ROM.
     
      2013 Subscription Members will automatically receive two copies of       this data. 2013 Standard Members may request a copy as part of       their 16 free membership corpora.  Non-members may license this       data for US$1000.
     
     
   

*

(3)Mixer 6
        Speech was developed by LDC and is comprised of 15,863 hours       of telephone speech, interviews and transcript readings from 594       distinct native English speakers. This material was collected by       LDC in 2009 and 2010 as part of the Mixer project, specifically       phase 6, the focus of which was on native American English       speakers local to the Philadelphia area.

The speech data in this release was collected       by LDC at its Human Subjects Collection facilities in       Philadelphia. The telephone collection protocol was similar to       other LDC telephone studies (e.g., Switchboard-2 Phase III Audio -       LDC2002S06): recruited
      speakers were connected through a robot operator to carry on       casual conversations lasting up to 10 minutes, usually about a       daily topic announced by the robot operator at the start of the       call. The raw digital audio content for each call side was       captured as a separate channel, and each full conversation was       presented as a 2-channel interleaved audio file, with 8000       samples/second and u-law sample encoding. Each speaker was asked       to complete 15 calls.

The multi-microphone portion of the collection       utilized 14 distinct microphones installed identically in two       mutli-channel audio recording rooms at LDC. Each session was       guided by collection staff using prompting and recording software       to conduct the following activities: (1) repeat questions (less       than one minute), (2) informal conversation (typically 15       minutes), (3) transcript reading (approximately 15 minutes) and       (4) telephone call (generally 10 minutes). Speakers recorded up to       three 45-minute sessions on distinct days. The 14 channels were       recorded synchronously into separate single-channel files, using       16-bit PCM sample encoding at 16000 samples/second.

The recordings in this corpus were used in NIST       Speaker Recognition Evaluation (SRE) test sets for 2010 and 2012.       Researchers interested in applying those benchmark test sets       should consult the respective NIST Evaluation Plans       for guidelines on allowable training data for those tests.

The collection contains 4,410 recordings made       via the public telephone network and 1,425 sessions of multiple       microphone recordings in office-room settings. The telephone       recordings are presented as 8-KHz 2-channel NIST SPHERE files, and       the microphone recordings are 16-KHz 1-channel flac/ms-wav files.    

Mixer 6 Speech is distributed on one hard       drive.
     
      2013 Subscription Members will automatically receive one copy of       this data on hard drive.  2013 Standard Members may request a copy       as part of their 16 free membership corpora. As a Members-Only       release, Mixer 6 Speech is not available for non-member licensing.

.