CSR '95 Hub-3 Development Test Data Read CSRNAB Multi-Microphone Recordings NIST Speech Discs 26-3.1, 26-4.1, 26-5.1 May, 1996 This corpus of read speech was collected by NIST to serve as development test data for the 1995 CSR Hub-3 tests. The data was collected using using the NIST-developed Multi-Microphone (MUM) Speech Data Collection System. The MUM system permits the the simultaneous collection of speech from eight microphones. These discs (26-3.1 - 26-5.1) contain multi-microphone speech for 20 subjects reading the 40 ARPA/Dragon adaptation sentences and unique sets of nominally 15 NAB 1994 H1 Development Test sentences. Background noise samples were also recorded before and after each data collection session. The data were collected in two noise environments (locations) with 10 of the subjects in each location. NOTE: If you intend to implement the 1995 ARPA CSR Hub-3 Tests, you should use this data in accordance with the Hub-3 test and scoring specifications on NIST Speech Disc 26-7.1 which contains the corresponding evaluation test material. The 20 subjects are apportioned across the 3 discs by recording location as follows: 26-3.1 - Location 1 Recordings: 15 NAB sentences & 40 adaptation sentences Subjects: 703-705, 707-70a, 70c, 70f, 70g 26-4.1 - Location 2 Recordings; 15 NAB sentences: Subjects: 70k, 70m-70o, 70q-70s, 70u-70w 26-5.1 - Location 2 Recordings; 40 adaptation sentences Subjects: 70k, 70m-70o, 70q-70s, 70u-70w The top-level directory of this disc contains the following files and subdirectories: discinfo.txt - listing of directories and files on this disc csr95/h3/ - directory containing Hub-3 corpora A more detailed description of the data collection setup and data is located in "csr95/h3/doc/h3_data.doc" A detailed description of the MUM data collection system is located in "csr95/h3/doc/mum.doc". The 1995 CSR Hub-3 test specifications are located in "csr95/h3/doc/et95h3sp.doc". For your convenience, the documentation under "csr95/h3/doc" has been copied to all three discs. Please see the "readme.doc" file under the "csr95/h3" directory to get started with the data.