This database is recorded to simulate the various environments by convolution of dry sources and numerrous impulse responses. Also, moving sound sources that can not be simulated by convolution of dry sources and numerous impulse responses are recorded in real environments. This database includes 9 impulse responses collected in 9 kinds of room, 5 kinds of actual recorded speech, 5 kinds of actual recorded talkers who were moving, etc.

Table 1 shows the instruments that were used for recording the data. Two variations of microphone arrays were employed.

Circle: Circle type microphone array (16 channels)

Cirline: Circle + Linear type microphone array (30 channels)

Figure 1 depicts the location of the transducers. Number 1 is the forward direction. A circle type microphone array has a diameter of 30cms, and has 16 transducers located at equal intervals. A linear type microphone array has 14 transducers located at 2.83cm intervals.

Figure 1: Transducer locations

Figure 2: Circle + Linear type microphone array

Performance

System Design : OPTOTRAK 3020 (by Nothern Digital Inc.)

Real time 3-dimensional position measuring system


1. Tracks 256 infrared diodes in real time


2. Three dimensional position real time measurement


3. Analyzing resolution performance of 0.01mm, RMS performance of 0.1mm


4. 400Hz sampling frequency at maximum


System Design : Data Acquisition Unit II (by Nothern Digital Inc.)


Synchronization signal input-output system


Record OPTOTRAK speech and position data in synchronization


1. Analog data : 16ch input and 2ch output


2. Digital : 8ch input



Figure 3 : OPTOTRAK

4. Recording conditions

The average ambient noise level, humidity and temperature in each recording room are shown in Table 2. Note that the ambient noise level is slightly higher for a moving talker because of the effects of OPTOTRAK. Symbols (*) indicates sound level meter ''under flow'' in the Table 2.

Table 2 : Ambient noise level, humidity and temperature

Room	Reverberation time	Temperature	Humidity	dBA	dBC
Anechoic room	0.0	21.4 degrees	44.4 %	15.9*	29.9
Echo room(cylinder)	0.12	-	-	18.5*	48.0
Echo room(cylinder)	0.31	-	-	18.5*	48.3
Echo room(cylinder)	0.38	-	-	17.4*	46.0
Echo room(panel)	0.3	22.1 degrees	37.1 %	18.9*	43.8
Echo room(panel)	1.3	22.1 degrees	37.1 %	19.6*	49.6
Conference room	0.78	20.7 degrees	36.2 %	43.2	50.9
Tatami - floored room (L)	0.60	11.5 degrees	37.0 %	44.7	55.1
Tatami - floored room (S)	0.47	12.0 degrees	36.8 %	44.8	56.7

5. Recording scenery

6. Recording configuration

Information from each experimental instrument




Anechoic room

Echo room




7. Recorded data

This database includes the 5 items in below. Impulse response is float type binary data (4 bytes) and speech data is short type binary data (2 bytes). Also, all binary types are little endian, and sampling frequency is 48kHz.

1. Fixed sound source

(A) Fixed impulse response *** Impulse responses are measured at several positions which are 2m distance from the microphone array.
(B) Fixed speech *** Speech signals are recorded forward of the microphone array at a distance of 2m.


2. Moving sound source

(A) Fixed impulse response *** Impulse responses are measured at several positions which are at a distance of 2m from the microphone array using a custom fabricated moving sound source system.
(B) Moving speech *** Speech signals are recorded using the custom fabricated moving sound source system (50 degrees to 130 degrees). The positions of the sound sources are recorded by OPTOTRAK.


3. Diffused sound environment

(A) Impulse response *** Impulse responses are measured by facing a loud speaker to wall.


4. Ambient noise

(A) Silence signal *** This is recorded for the purpose of adding ambient noise after the convolution of a dry source and an impulse response in order to reproduce a real acoustic signal.


5．Piston phone

(A) Transducer calibrated signal *** Each transducer is calibrated with a piston phone.


Impulse response
Impulse response is measured using the TSP method [1]. TSP length is 65536 points. The number of synchronous addition is 16 times. TSP used during recording and time inversed TSP are shown below. The figure shows a sample of impulse response.
[TSP data], [Time inverse TSP data]


A sample of impulse response

Speech data
Speech data is recorded for 90 sentences that include phoneme balanced 50 sentences * 1 subject in ATR (Advanced Telecommunications Research Institute International ) database and 10 sentences * 4 subjects in TIMIT (The DARPA TIMIT Acoustic-Phonetic Continuous Speech Corpus）database. Speech data is compressed by "ZIP." If you conduct "UNZIP," speech data appears in each file (mmysda01.1, mmysda01.2,...） in the directory. Samples of speech waves which are recorded in an anechoic room and in a variable echo room are shown below.

Anechoic room

Variable echo room

Moving sound source
Moving sound sources are recorded with the custom fabricated moving sound source system as shown in Figure 2. The microphone array is placed at the base of a center pole after fixing the speaker at the end of rod as shown in Figure 2. Impulse responses for fixed positions as shown in above 2.(A) are also recorded at some points on this track. Speech signals from a moving sound source are recorded by manually moving the rod. Thus, the source positions which are measured by OPTOTRAK are also recorded each time. Source position data format is little endian float type binary data (4 bytes) and 100Hz sampling. File; *.x, *.y and *.z show the coordinates that represent the distances from the microphone array to the speaker in units [mm]. The figure below shows the track of moving sound recorded with the custom fabricated moving sound system.

Custom fabricated moving sound system

Locus of moving source

Diffused sound environment
Impulse response was measured by turning a speaker towards the wall in the diffused sound environment. Recording configurations are shows below.

Ambient noise
Ambient noise is recorded 10 times * 2 subjects at conference room.

Piston phone
Transducer calibration signals were recorded using a piston phone (ONSOKU TYPE2126).

8. A list of data

Notes of notation

RT:        Reverberation time in room
CH:        The number of transducers 
ang:       The number of recorded directions
len:       The length of data per transducer (in speech data, the number of sentences)
Byte:      Byte order of files (all data is little endian)
Total:     Capacity 


                         Room                           RT      CH      ang     len     Byte    Total

/MICARRAY/data1 ... Impulse response
  +--/circle                                                            
  |    +--- ANE/imp000   Anechoic room                  0.00    16      36      5000    4       11.2MB
  |    +--- E1A/imp000   Echo room(cylinder)A           0.12    16      1       12000   4       753kB
  |    +--- E1B/imp000   Echo room(cylinder)B           0.31    16      19      30000   4       34.9MB
  |    +--- E1C/imp000   Echo room(cylinder)C           0.38    16      19      36000   4       41.8MB
  |
  + --/cirline
       +--- E2A/imp000   Echo room(panel)A              0.30    30      9       30000   4       31.0MB
       +--- E2B/imp000   Echo room(panel)B              1.30    30      9       125000  4       128.8MB
       +--- OFC/imp000   Conference room                0.78    30      9       75000   4       77.3MB
       +--- JR1/imp000   Tatami - floored room (L)      0.60    30      9       60000   4       61.9MB
       +--- JR2/imp000   Tatami - floored room (S)      0.47    30      7       45000   4       36.1MB


/MICARRAY/data2 ...Speech( E2A in DVD-ROM Vol.1, and E2B OFC JR1 JR2 in DVD-ROM Vol.2)
  +-- /cirline/
       +--- E2A/mmysda01 Echo room(panel)A              0.30    30      1       90sent. 2       1.2GB                 
       +--- E2B/mmysda01 Echo room(panel)B              1.30    30      1       90sent. 2       1.2GB
       +--- OFC/mmysda01 Conference room                0.78    30      1       90sent. 2       1.2GB
       +--- JR1/mmysda01 Tatami - floored room (L)      0.60    30      1       90sent. 2       1.2GB
       +--- JR2/mmysda01 Tatami - floored room (S)      0.47    30      1       90sent. 2       1.2GB


/MICARRAY/data3 ...Movement (Impulse response)
  +-- /cirline
       +--- ANE/imp000   Anechoic room                  0.00    30      5       5000    4       2.9MB
       +--- E2A/imp000   Echo room(panel)A              0.30    30      5       30000   4       17.2MB
       +--- E2B/imp000   Echo room(panel)B              1.30    30      5       125000  4       71.6MB
       +--- OFC/imp000   Conference room                0.78    30      5       78000   4       43.0MB
       +--- JR1/imp000   Tatami - floored room (L)      0.60    30      5       60000   4       34.4MB


/MICARRAY/data4 ...Movement(Speech)(in DVD-ROM Vol.3)
   +-- /cirline
       +--- ANE/mmysda01 Anechoic room                  0.00    30      1       90sent. 2       1.2GB
       +--- E2A/mmysda01 Echo room(panel)A              0.30    30      1       90sent. 2       1.2GB
       +--- E2B/mmysda01 Echo room(panel)B              1.30    30      1       90sent. 2       1.2GB
       +--- OFC/mmysda01 Conference room                0.78    30      1       90sent. 2       1.2GB
       +--- JR1/mmysda01 Tatami - floored room (L)      0.60    30      1       90sent. 2       1.2GB


/MICARRAY/data5 ...Diffused sound environment
   +-- /cirline
       +--- OFC/imp_rev Conference room                 0.78    30      1       75000   4       8.6MB


/MICARRAY/data6 ...Ambient noise
  +-- /cirline
       +--- OFC/ambient1 Conference room                0.78    30      1       480000  2       27.5MB
       +--- OFC/ambient2 Conference room                0.78    30      1       480000  2       27.5MB


/MICARRAY/data7 ...Piston Phone
   +--/cirline
       +--- Equalize     -                              -       30      1       144000  2       8.3MB

 

Reference

[1]Y. Suzuki, F. Asano, H.Y. Kim, and T. Sone,
"An optimum computer-generated pulse signal suitable for the measurement of very long impulse responses,"
J. Acoust. Soc. Am., Vol. 97, No. 2, pp. 1119-1123, 1995.




[Back to top of RWCP Sound Scene Database ]

---

RWCP Sound Scene Database in Real Acoustical Environments
Copyright (c) 1998-2001 Satoshi Nakamura, and Takanobu Nishiura, ATR Spoken Language Translation Research Laboratories.