Mobile Platform Acoustic-Frequency Environmental Tomography

From SpeechWiki

Revision as of 19:27, 30 May 2009

Compute room geometry and mic position

Room, loudspeaker, mic.

Mic and speaker unmoving, known distance apart.

Play chirp or sine sweep. Not MLS: too slow.

From recorded sound, estimate room's geometry w.r.t. mic and speaker.

• Lae-Hoon's master's thesis has an algorithm for this.
• Verify this algorithm against plywood-cube chirp recordings.
• Generalize to non-shoebox rooms.
• Generalize to a dynamic algorithm for a moving mic and speaker (robot).
• Generalize to a changing room shape.

Space-mapping robot

EMAR, Expendable Mapping Acoustic Robot

Put a speaker near a microphone, to map a space.

Sine sweep or chirp, not MLS. We need speed and reflector attributes, not precision.

• Fast mode. Catch the first two or three echoes and find out where the two nearest surfaces are, and match those against things on the video camera in order to determine space geometry
• Slow mode. Measure the detailed room response at a few different locations (by moving the microphone), use this information together with video (hybrid, like AVSR, increases accuracy) to build up and test hypotheses for the room geometry.

Simulator (ruby, glut).

Vehicle

Prototype: Illy II, an Arrick Trilobot. In BI 1510 (locked). Contacts Logan Niehaus <niehaus4@illinois.edu>, Stephen Levinson <sel@ifp.uiuc.edu>.

Summer, don't add hardware. Use its two mics and ADCs, 802.11b, low-level API (write ruby bindings?). Replace its speech-synth loudspeaker with a more linear piezo? It has 6 drive motors. Its 14 sensors include:

• whiskers for collision detection
• temperature+humidity to compute local speed of sound
• position: laser rangefinder, ultrasonic rangefinder (untested), odometry (inaccurate).

Fall, use more of its 8 channels of ADC for a tetrahedral mic array. Mount mics on pan-tilt head, so head-rotation verifies the array's angular accuracy. Loosely couple the software connecting payload to vehicle.

Outdoors: Conventional two-tread "tank".

Payload

What's the weight, size, power, and cooling requirements of payload components?

• 2 mics
• 1 speaker
• power amplifier
• computer handling mics + speakers
• computer running Lae-Hoon's algorithm

How much computation happens on the robot, and how much on its base-station laptop?

• How robust and wide is the data path between them?
• How short a battery life can we tolerate?
• Can a fast onboard CPU run cool enough? (ammonium nitrate + water first aid "instant ice pack", shaken by robot when it feels hot)

Application

Small robot rolls ahead of firefighters into a collapsing building, and maps it to reduce the risk they are exposed to. Small lets it reach places inaccessible to humans. CRASAR recommends only high-level commands given by human operator.

Related NSF award.

Training venue: Disaster City.

IFSI

Test in "Collapse Street" collapsed building simulators of IFSI. After Illi/Norbert in-the-lab study, contact IFSI's Gavin Horn 265-6563 <ghorn@illinois.edu> about collaborating, designing an outdoor prototype, and applying for funding.

Treads should succeed in IFSI's rubble, because it's not sand or mud or wet leaves.

Acoustic environment: nonstationary noises, like campfire crackle and water hoses. Many fast chirps tolerate such noise?

Passive mic could listen to crackles to guess wall locations, if crackle and wall correlate.

Robot lightweight enough for a firefighter to throw through a door or over an obstacle. Ingress faster than exploration.

MIRV. Launch like a mortar, perhaps just by dropping into the path of the water hose. In flight, compressed springs release by electric wires burning through, to scatter robots (which themselves MIRV, 2 or 3 stages). Robots chirp in flight, while bouncing, while at rest. Learn a space's rough geometry within 5 seconds, improving accuracy thereafter. Battery life of smallest bots need not exceed 60 seconds (capacitors for prototypes).

Two extensions of Lae-Hoon's Jan 30 paper review

1. Remove assumption of time invariance of RIR, because listeners' heads and ears move enough to degrade performance at high frequencies.

2. Extend their simulation to experiment with real microphones.

Of each mic in an array:

• nonuniform frequency response
• nonuniform spatial ("off-axis") response
• nonuniform accuracy of measurement of spatial position
• nonuniform accuracy of measurement of orientation, if mic isn't "omnidirectional"
• nonuniform SNR
• correlated inter-mic noise (not independent Gaussians) from multichannel preamplifier
• actual crosstalk between channels, again from preamp
• noises in domains other than amplitude-vs-time

At some point, even if mics cost no money, these inaccuracies suggest that adding mics would degrade rather than improve performance.

Sensitivity analysis of these things could be done entirely in simulation, as a quickly publishable result. A second paper exe tests that with experiments.

Later work

• For more accuracy, estimate nonlocal speed of sound from computed and remembered values.
• Secondary computation: ASR for "help!" and screams. Tiny vocabulary. Robust to background noise.
• Flock of robots. Faster, but tricky crosstalk.

Where we're at, 2009 Feb 27

We have Sarah's speaker-to-mic recordings, dimensions/positions of room, mics, speakers:

• raw .wav files and deconvolved .mat files
• MLS and chirp deconvolutions
• from each of 4 speakers, to each of 40 mic positions
• from some speaker-pairs, to each of 24 mic positions

Speaker-pair recordings are incomplete (only 4 of 6 possible pairs). But we could use them as sanity checks on the single-speaker recordings, instead of as primary data.

The plywood cube (actually particleboard with 2x4 framing) has been demolished. The thin-glass parts of the speakers have been demolished.

ISL still has the amplifiers, speaker drivers, and mics. One of the two Earthworks omnidirectional mics is malfunctioning and needs replacing, if we need stereo recording.

ISL's multichannel recording PC, fruitfly.isl.uiuc.edu, has moved south with its 8-channel i/o interface.

If we reconstruct a plywoodcube, prefer flush-with-wall conventional speakers over the original motivation of glass-speakers-through-cubewall-slits.

Abandoned ideas

Corpus

To validate room response models. No more "research." Mention image-source and other algorithms.

Refine image-source

Frequency-dependent wall reflection and/or air transmission, and other subtle refinements as the data suggests. Look at CATT and other commercial packages for architectural acoustics; they include, e.g., hybrid image source/ray-tracing room responses, with frequency response of different materials implemented at each reflection.

In early 2008, Mark guessed at least 12 months until "good-sounding" room inverse (40 dB, not just Bowon's 10 dB) in simulation, warranted before sawing particleboard.

Mask the reverberant tail with 10 dB SNR noise, since later echos may overlap too much to cancel rigorously.

Validate room response models

Play sounds convolved by the plywoodcube's computed inverse-impulse-response. Compare the recorded results to the original unconvolved sounds. In simulations, or with a fresh plywoodcube.

A wood "phonebooth" would fit almost anywhere. A larger phonebooth at ISL, operated remotely since it's beyond walking distance.