This directory contains multi-microphone data recorded by Tom Sullivan (tms@cs.cmu.edu) at Carnegie Mellon University. All data are sampled at 16 kHz, 16-bit linear sampling. There are 3 directories included: 15element -- Recorded at Carnegie Mellon University. 8element -- Recorded at Carnegie Mellon University. rutdata -- Recorded at Rutgers University. The contents of each directory are described individually in more detail below. 15element --------- These utterances were collected with a 15-element array. The array spacing is such that it is actually three 7-element sub-arrays (with different spacing) interleaved (ala. Jim Flanagan's array at ATT/Rutgers U.) Some elements are shared between the sub-arrays. If the minimum spacing of this array is N-cm. The array looks roughly like this: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Where element 8 is in the exact center of the array. Elements 5, 6, 7, 8, 9, 10, and 11 are a 7 element array with a spacing of N cm between elements. Elements 3, 4, 6, 8, 10, 12 and 13 are a 7 element array with a spacing of 2*N cm between elements. Elements 1, 2, 4, 8, 12, 14 and 15 are a 7 element array with a spacing of 4*N cm between elements. Of course you can feel free to combine the elements in any manner you desire for your own experiments. We used them to study different element spacing from data that was collected simultaneously to each of the sub-arrays. Within this directory are files of the form: {utterance}-{subject}-{microphone_type}.adc {utterance} is either an* or cen*, where "*" is a number. These are alphanumeric and census utterances in the AN4 dataset used often here at CMU. {subject} is the same 4-letter code used to name the sub-directories. {microphone_type} is the type of microphone used and a key into the experimental conditions. "senn" is the Sennheiser HMD414 headset closetalking microphone, used in every set as a control. "arr" is a microphone array having multiple elements. All array elements are Panasonic WD-063 noise cancelling electret condenser elements. There are 6 data sets collected with this array: 1) This set was collected in a noisy computer lab at Carnegie Mellon Univ. with an array of 15-elements with a minimum spacing of 3 cm. It is denoted by "arr3A" for the 15 array elements and "senn3" for the closetalk. The subject sat one meter from the center of the array. Ex: an101-mtms-arr3A.adc (15-channel array, 3 cm minimum spacing) an101-mtms-senn3.adc (closetalking control signal for above) 2) This set was collected in the same noisy lab as above but with the 15-element array with a minimum spacing of 4 cm. It is denoted by "arr4A" and "senn4" for the closetalk. The subject sat one meter from the center of the array. Ex: an101-mtms-arr4A.adc (15-channel array, 4 cm minimum spacing) an101-mtms-senn4.adc (closetalking control signal for above) 3) This set was collected in a conference room with the 15-element array with a minimum spacing of 4 cm. The conference room is larger than the noisy lab, but didn't have all of the computer fans. It is denoted by "arrC1A" and "sennC1" for the closetalk. The subject sat one meter from the center of the array. Ex: an101-mtms-arrC1A.adc (15-channel array, 4 cm minimum spacing, 1 meter dist.) an101-mtms-sennC1.adc (closetalking control signal for above) 4) This set was collected in the same conference room as 4) above with the 15-element array with a minimum spacing of 4 cm. It is denoted by "arrC1A" and "sennC1" for the closetalk. The subject sat three meters from the center of the array. Ex: an101-mtms-arrC3A.adc (15-channel array, 4 cm minimum spacing, 3 meter dist.) an101-mtms-sennC3.adc (closetalking control signal for above) 5) This set was collected the same conference room as above with the 15-element array with a minimum spacing of 4 cm, but also had an AM talk-radio jamming signal at approximately 45 degrees off-axis from the center of the array, competing with the speaker. It is denoted by "arrCR1A" and "sennCR1" for the closetalk. The subject sat one meter from the center of the array. Ex: an101-mtms-arrCR1A.adc (15-channel array, 4 cm minimum spacing, 1 meter dist., radio jamming signal) an101-mtms-sennCR1.adc (closetalking control signal for above) 6) This set was collected the same conference room as above with the 15-element array with a minimum spacing of 4 cm, but also had an AM talk-radio jamming signal at approximately 45 degrees off-axis from the center of the array, competing with the speaker. It is denoted by "arrCR1A" and "sennCR1" for the closetalk. The subject sat three meters from the center of the array. Ex: an101-mtms-arrCR3A.adc (15-channel array, 4 cm minimum spacing, 3 meter dist., radio jamming signal) an101-mtms-sennCR3.adc (closetalking control signal for above) 8element -------- There are ten "subject" sub-directories included. Their names are each four letters long. The first letter denotes the gender of the speaker ("m" or "f"). All data we've collected thus far used only male speakers. The final three letters are the initials of the subject. Hence "mtms" is a male speaker with the initials TMS. Within each subject's sub-directory are files of the form: {utterance}-{subject}-{microphone_type}.adc {utterance} is either an* or cen*, where "*" is a number. These are alphanumeric and census utterances in the AN4 dataset used often here at CMU. {subject} is the same 4-letter code used to name the sub-directories. {microphone_type} is the type of microphone used and a key into the experimental conditions. "senn" is the Sennheiser HMD414 headset closetalking microphone, used in every set as a control. "arr" is a microphone area having multiple elements. All array elements are Panasonic WD-063 noise cancelling electret condenser elements. There are 3 different data sets contained in these directories. They are listed as follows: This set contains 10 male speakers each speaking 14 utterances. The microphone array used had eight (8) elements, and these utterances are denoted by "arrA" as the {microphone_type}. The 8 elements were spaced linearly and with a spacing of 7 cm between elements. The subject sat directly in front of the array at a distance of 1 meter from the center. A pair of Crown PZM6FS microphones were also used to collect a stereo pair of Crown PZM signals to compare the performance of a quality set of omnidirectional microphones to the Panasonic WD-063 array elements. The closetalking control signal is denoted by "senn" only. The set was collected in a noisy computer lab at Carnegie Mellon Univ. with many computer and disk-drive fans. Ex: an101-mtms-arrA.adc (8-channel array) an101-mtms-pzmS.adc (stereo pair of PZM6FS microphones) an101-mtms-senn.adc (closetalking control signal) rutdata ------- These data were collected at the CAIP Center at Rutgers University in 1991 with the help of Jim Flanagan and Joe French. The experimental setup consisted of two 23-element microphone arrays built at ATT Bell Labs, a Crown PZM6FS microphone, and a Sennheiser HMD414 head- mounted closetalking microphone. The arrays had 23 microphone elements each (comprising three interleaved arrays of 11-elements each). One of the arrays had a bandwidth of 8kHz (standard DARPA speech bandwidth) and the other a bandwidth of 4kHz (standard telephone speech bandwidth). There are two sub-directories: 1meter and 3meters. These directories refer to the distance of the speaker from the center of the microphone array. Within each directory are sub-directories, one for each speaker. Within each subject's sub-directory are files of the form: {utterance}-{subject}-{microphone_type}.adc {utterance} is either an* or cen*, where "*" is a number. These are alphanumeric and census utterances in the AN4 dataset used often here at CMU. {subject} is the same 4-letter code used to name the sub-directories. {microphone_type} is the type of microphone used and a key into the experimental conditions. "sen" is the Sennheiser HMD414 headset closetalking microphone, used in every set as a control. "pzm" is a Crown PZM6FS microphone, "arrA(1m or 3m)" is a 8kHZ bandwidth array described above, and "arrB(1m or 3m) is a 4kHZ bandwidth array as describe above. All array elements are Panasonic WD-063 noise cancelling electret condenser elements. The arrA and arrB files in this data set are monophonic files. They are created by combining the array element outputs in real-time via hardware via delay and sum beamforming. ------------------------ The sentence transcripts for each of the files in "15element" and 8element" are contained in the "transcripts" directory. The sentence transcripts for each of the files in "rutdata" are contained in the "rutdata/transcripts/trans{1m,3m}" directories. ------------------------ The file ad.h in this directory is a C "include" file that contains the structure of the soundfile headers found on the .adc files.