Treble simulation of medium-sized room (CR2 from BRAS)
Treble simulation of medium-sized room (CR2 from BRAS)
test
ABSTRACT
Treble outperforms any conventional geometrical acoustics (GA) software in simulating single reflections (BRAS - RS1), diffraction from a finite object (BRAS - RS6),diffraction from a large barrier (BRAS – RS6) by directly solving the wave equation This study shows Treble’s performance in a medium sized seminar room, namely, BRAS – CR2.
1. Introduction
Accurate room acoustic simulations fundamentally rely on accurate simulation of individual reflections off the surfaces, diffraction and scattering due to obstacles and furniture. It has been shown that Treble accurately captures the reflections and diffraction for various conditions, BRAS RS1, RS5, RS6, etc. To validate a Treble simulation of a room, BRAS scene CR2 is chosen. The room is basically rectangular, 8.4 m ×6.7 m × 3 m. The surfaces are rather reflective as shown in Table 1.
Figure 1. CR2 and Treble simulation snapshot [3].
2. Measurement setup
A 3-way dodecahedron sound source was used during the measurement and an omni-directional microphones B&K 4134 was used in the seminar room. Similarly, the Treble simulation uses an omni-directional source and omni-directional microphone in the simulation. The transition frequency is 710 Hz, meaning the 500 Hz octave band is simulated by DGFEM. The locations of the mid-frequency sources are LS1 (0.93, -2.55 0.72) m and LS2 (0.19, 2.88, 0.72) m. The high-frequency source has a height of 1.023 m while the low source has a height of 0.462 m. MP1 is located at (-0.99, 1.46, 1.23) m, MP2 at (0.44, -0.147, 1.23) m, MP3 at (1.36 -0.6, 1.23) m, MP4 at (-1.11, -0.26, 1.23) m, MP5 at (-0.99, -1.41, 1.23) m. The absorption coefficients in the BRAS database (fitted absorption) are shown as follows:
Table 1. The absorption coefficient from BRAS database
Two transfer function comparisons for LS1MP4 and LS2MP5 are shown in Fig. 2. There is a reasonable agreement in predicting the room modes at lower frequencies below 200 Hz. One can see slightly more sharp peaks from Treble, indicating that the damping in the Treble simulation could be underestimated.
Figure 2. Transfer function comparisons. Upper: LS1MP4, Lower: LS2MP5.
The ISO 3382 parameters are compared in Fig. 3. The ISO 3382 parameters are calculated from 2 sources and 5 receiver locations (thus 10 combinations) and then their averages and standard deviations are plotted. Frist, EDT is a reverberation parameter which is really sensitive to early reflections. EDT is a more difficult parameter to make a good agreement between simulations and measurements. Typically the spatial deviation of EDT is larger at lower frequencies, which is not well captured by GA simulations. Treble results shown in Fig. 3 agree well with the measurement not only the average values but also the spatial standard deviations. Center time, Ts, is also a sensitive parameter to the temporal structure of the impulse responses and known to be more difficult to match than C50 or D50, which have a fixed demarcation point separating the early and late reflections. The average values of Ts from Treble are reasonable as well as the standard deviation. C50 is a typical parameter that quantifies the speech clarity. C50 agrees well, particularly the average values. D50 is another clarity parameter that uses 50 ms as the transition time from early to late reflections. We can observe good agreements of D50 parameter between the simulation an BRAS measurement.
Figure 3. ISO 3382 parameter comparisons. EDT, Ts, C50 and D50.
3. Conclusions
This study investigates how a Treble simulation is performed in a room condition, typical classroom or seminar room condition. Treble’s acoustic parameter predictions agree well with the measurement data thanks to its accurate acoustic solver. A source of the errors are uncertainties in the source and receiver locations in the measurement, particularly the measurement used a 3-way dodecahedron source. Despite all the uncertainties, the Treble simulation results are convincing over a large frequency range from 63 Hz to 4 kHz.
References
[1] Benchmark of Room Acoustical Simulation (BRAS) database CR2: https://depositonce.tu-berlin.de/bitstreams/53c3cf64-3547-4aa6-946b-1b4755729f2a/download