meeting notes

3/12

Project Goals
- Create an artificial room and simulate sound from different sources that can move and play simultaneously
- Split audio into LR stereo for accurate binaural sound reproduction
- Model walls as filters that change based on material
- Modify sampling rate to see if it affects fidelity (e.g. slow sampling leads to missing an echo)
- Parametrize the size of the room, locations of the sounds and listener, materials, objects in room, volume (stretch)
Applications of sound synthesis/simulation
- Home entertainment/VR
- Music
- Theater/room design
  - Beneficial to simulate a space before you build it
- Using the knowledge of room effects to isolate sound from the environment
  - E.g. making sound underwater easier to hear by removing the water effects
Metrics
- Efficiency of algorithm (can we compare this to anything?)
- Compare simulated sound to real sound by modeling a real room and recording in that room
Point targets
- Creativity + Application
  - Demonstrated through demos and complexity of sound spaces (e.g. a box in the middle of the room is creative)
- Effort + Algorithms
  - Demonstrated through demos, versatility (parameterized), and speed
- Understanding + Application
  - Demonstrated through description of applications, and metrics
Next steps
- Everyone find one paper about sound simulation, sound-spaces, modeling, or materials (walls as systems). Write a short summary on how it is useful for the project.
- Get (at least) two mics and balloons for recording impulse responses
  - Goal: Measure the effect of wall material on impulse response (use a spectrogram to narrow-in around first echo)
  - Write summaries of the effects of each wall and theorize why they are that way
- Write a summary for the given room-modeling paper

3/24

Data Collection
- Because we are not sure if we’re interpreting the data correctly…
- Use existing data to make a prediction about what new data will look like (e.g. if we think we know the first reflection’s time delay and amplitude at 6 ft, extrapolate what it will be at 12 feet, then go collect data and verify)
- Get together (Tue or Wed?)
  - Bring mics and tape measure
  - If we can’t do this before Friday, it becomes a checkpoint on our timeline
- Sound sources: guitar?
  - Obviously not omnidirectional, but we’ll assume they are, and if we’re wrong and it sounds bad, we’ll change it
Algorithm Design
- Two approaches:
  - Create a ton of rays and track where they go
  - Calculate which rays will go to the receiver directly
  - Compare the two for performance
- Progress
  - Started putting in loops to go through rays and find collisions with all walls
  - Nice parameterized framework to change size of room and location of source and receiver
- Challenges
  - More wall collisions than is possible. For example, for 8 rays, we’re getting 9 or 10 collisions.
  - Key: This is repeatable; for less than 8 rays, the number of collisions always equals the number of rays. For 8, it’s always 8-10.
  - Precision error: Had to add feature to check if a calculated wall collision was within TOLERANCE distance of the real wall (e.g. the collision would be at 2e-10 instead of 0)

3/27

Code Progress
- Source of collision point bug from last time: Calculation of D values of plane equations needed negative sign.
  - Because this was missing, the walls weren’t where we intended them to be.
- We’ll assume specular reflection.
- Do some math to figure out CROSS_TOLERANCE

4/9

Project Discussion Planning
- Topic: Sound synthesis
- Project: Place a source and receiver in an imaginary room of arbitrary size and with walls of different materials, and generate what the receiver would hear from the source. Demonstrate how changing the room size and material types changes the sound.
- Progress:
  - Algorithm to compute the delay and attenuation of n-th order echoes using wall absorption coefficients, and generate a single room impulse response
  - When convolved with Bach audio, you can hear the effect of the room (e.g. reverberation)
  - Algorithm to find wall absorption coefficients from collected impulse response data
  - Data set of HRTF data that we will run through Matlab’s interpolation function to generate a HRTF
  - A failed ray-tracing algorithm to try to compute echoes
- Which rubric strengths we are targeting and how we’re targeting them:
  - Creativity+Application: Demonstrated through demos and complexity of sound space (different materials, sizes)
  - Effort+Algorithms: Demonstrated through versatility of our algorithms and study of absorption
  - Understanding of Application: Demonstrated through description of applications, and discussion of best approach (e.g. ray tracing failed)
- Intended DSP tools:
  - In-class: Linearity and the impulse response
  - In-class: Filtering
  - Out-of-class: Finding n-th order echoes in collected data, HRTF interpolation, ray model for sound
- A list of questions in order of priority
  - Q1: Are these deliverables enough to earn a high score, or do we need more?
  - Q2: How much data do you want to see in the report, or should we just list conclusions?
  - Q3: Do you want us to try to embed the code in the website, or only show the results?
  - Q4: Do you have any recommendations for procedures we could use to classify wall materials?
- Next steps (just for us)
  - Collect more wall data and process it to get absorption coefficients
  - Compare these with external data set
    - Carpet
    - Ceiling tile
    - Floor tile
    - Brick
    - Glass
  - Look into identifying a wall material from its frequency profile
  - Include sound decay with distance? 1/r
    - http://www.sengpielaudio.com/calculator-distance.htm