Skip to content

matlab/matlab-agentic-toolkit / skills-catalog/signal-processing/matlab-evaluate-acoustic-metrics

matlab-evaluate-acoustic-metrics

Verified

Select and use Audio Toolbox acoustic, psychoacoustic, and speech quality metrics. Use when computing loudness, sharpness, roughness, fluctuation strength, noise criteria (NC/RNC/RC), speech transmission index (STI), speech intelligibility index (SII), speech quality (ViSQOL, STOI), SPL, reverberation time (RT60), tone-to-noise ratio, or prominence ratio. Also use when asked about PESQ, POLQA, calibration factor, EBU R 128, LUFS, HVAC noise, background noise evaluation, or how to measure audio quality or noise compliance. Covers calibration, standard selection, signal alignment, and the complete function inventory including what does NOT exist.

references/spl-measurement.md

View on GitHub
# SPL Measurement

Guidance for selecting parameters, interpreting results, and guiding user workflows for `splMeter`.

## Intake Questions

### 1. What is the measurement purpose?

| Purpose | Frequency Weighting | Time Weighting | Key Output |
|---------|-------------------|----------------|------------|
| Occupational noise exposure (hearing protection) | A-weighting | Slow | Leq (8-hour TWA) |
| Environmental noise monitoring | A-weighting | Fast | Leq, Lmax |
| Peak assessment (impulse hazard) | C-weighting | — | Lpeak |
| Product noise testing (to spec sheet) | A-weighting | — | Leq |
| Machinery noise for NC/RC compliance | Z-weighting | Fast | Octave-band Leq |
| Regulatory environmental limit | Depends on regulation | Depends on regulation | Varies |
| Building acoustics (background noise) | Z-weighting | Fast | Octave-band Leq → pass to `noiseCriteria` |
| General "how loud is it?" | A-weighting | Fast | Leq |

### 2. Frequency weighting selection

| Weighting | What it does | When to use |
|-----------|-------------|-------------|
| A-weighting | Approximates human hearing sensitivity at moderate levels; attenuates low frequencies | Most general-purpose measurements, occupational noise, environmental noise, product noise specs |
| C-weighting | Flatter response; mild LF roll-off below 31.5 Hz | Peak measurements (Cpeak), high-level noise (>100 dB), assessing low-frequency content, some regulatory requirements |
| Z-weighting | Flat (no filtering) — raw acoustic pressure level | Octave-band analysis for NC/RC/RNC, feeding into compliance functions, when you need unweighted per-band levels, research |

**Guidance:**
- If the user says "dB" without qualification, they likely mean dB(A) — confirm
- If they need octave-band levels for noise compliance, use Z-weighting (the compliance functions apply their own corrections)
- C-weighting for peak is standard in occupational noise (IEC 61672 specifies C-peak for impulse hazard)
- A-weighting underestimates the perceived loudness of low-frequency noise — if the user complains about "rumble" but A-weighted levels seem low, suggest octave-band analysis or C-weighting

### 3. Time weighting selection

| Time Weighting | Time Constant | Behavior | When to use |
|----------------|---------------|----------|-------------|
| Fast | 125 ms attack/release | Tracks fluctuations; standard for most measurements | General purpose, environmental monitoring, product testing |
| Slow | 1000 ms attack/release | Smoothed; averages short-term fluctuation | Occupational noise exposure, stable environments, when fast readings fluctuate too much to read |
| Custom | User-defined | For specialized standards | Rarely needed; specific regulatory or research requirements |

**Guidance:**
- Default to Fast unless the user has a specific reason for Slow
- If levels fluctuate rapidly and the user wants a single characterization, Leq (equivalent continuous level) is more meaningful than any single time-weighted reading
- For impulsive noise (hammering, gunfire), neither Fast nor Slow captures the peak — use Lpeak output

### 4. Bandwidth selection

| Bandwidth | Output columns | When to use |
|-----------|---------------|-------------|
| `"Full band"` | 1 (overall level) | Quick overall SPL, occupational noise, simple monitoring |
| `"1 octave"` | 1 per octave band | NC/RC/RNC input, spectral diagnosis of noise sources, building acoustics |
| `"1/3 octave"` | 1 per 1/3-octave band | Detailed spectral analysis, product noise standards, transmission loss |
| `"2/3 octave"` | 1 per 2/3-octave band | Rarely used; specific legacy standards |

**Guidance:**
- If the goal is noise compliance (NC/RC/RNC), use `"1 octave"` with Z-weighting — the compliance functions expect octave-band Leq
- If the goal is a single number (occupational, environmental), use `"Full band"`
- If the user wants to "see what frequencies are loud," use `"1 octave"` or `"1/3 octave"`

### 5. Calibration

Ask about the full signal chain:
- **Recording device:** What mic/interface was used? What is the sensitivity?
- **Calibration method:** Did they use a pistonphone or physical SPL meter for system calibration?
- **CalibrationFactor:** Do they already have one? If not, see the Calibration section in SKILL.md

If calibration is unknown and absolute levels are needed, the measurement is unreliable. Recommend:
1. Re-record with a calibration reference, OR
2. Use `calibrateMicrophone` with a known-SPL reference tone recording

For *relative* comparisons (before/after, product A vs B), calibration matters less as long as the same setup is used for both recordings.

### 6. Multichannel input

`splMeter` accepts matrix input where each column is an independent channel. For multichannel:
- `CalibrationFactor` can be a vector (one value per channel)
- Outputs become 3D arrays when using octave/fractional-octave bandwidth: L-by-B-by-C (samples × bands × channels)
- Each channel is metered independently

## splMeter vs poctave

Both `splMeter` and `poctave` compute octave-band levels. They produce identical results when filter orders match:

```matlab
% These are equivalent:
spl = splMeter(SampleRate=fs, Bandwidth="1 octave", FrequencyWeighting="Z-weighting", ...
    OctaveFilterOrder=2, CalibrationFactor=1, TimeInterval=duration);
[~, Leq] = spl(x);
levels_splMeter = Leq(end,:);

[p, cf] = poctave(x, fs, "power", BandsPerOctave=1, FilterOrder=2);
levels_poctave = 10*log10(p' / (2e-5)^2);
```

The default filter orders differ — `splMeter` uses `OctaveFilterOrder=2`, `poctave` uses `FilterOrder=6`. This causes ~1–2 dB differences if not matched.

### TimeInterval and Leq accumulation

`splMeter` resets its Leq accumulation at each `TimeInterval` boundary. If the signal length is not an integer multiple of `TimeInterval`, the last interval is shorter. To compute overall Leq from per-interval values, duration-weight the energy sum:

```matlab
% Wrong: treats partial last interval as a full interval
Leq_wrong = 10*log10(mean(10.^(Leq_intervals/10)));

% Correct: weight by actual interval duration
Leq_correct = 10*log10((1/T_total) * sum(t_i .* 10.^(L_i/10)));
```

To avoid this issue entirely, set `TimeInterval` to the full signal duration — then `Leq(end,:)` is the overall Leq directly.

## Interpreting SPL Results

### Overall Sound Pressure Levels

| dB(A) | Environment / Source | Subjective Impression |
|-------|---------------------|----------------------|
| 20–30 | Quiet bedroom, recording studio | Very quiet; barely audible |
| 30–40 | Quiet library, rural ambient | Quiet; comfortable |
| 40–50 | Quiet office, residential area at night | Moderate; noticeable |
| 50–60 | Normal conversation at 1 m, office | Clearly audible; may interfere with concentration |
| 60–70 | Busy restaurant, TV audio | Loud; raised voice needed for speech |
| 70–80 | Vacuum cleaner, busy traffic | Very loud; prolonged exposure fatiguing |
| 80–85 | Power tools, loud music | Hearing protection threshold (occupational) |
| 85–90 | Heavy machinery, motorcycle | OSHA action level; hearing damage begins |
| 90–100 | Lawn mower, band rehearsal | Painful without protection |
| 100–120 | Chainsaw, rock concert, jet takeoff at distance | Immediate hearing damage risk |
| >120 | Threshold of pain, close to explosion | Physical pain; acoustic trauma |

### Occupational Noise Exposure Limits

| Standard | Permissible Exposure Limit (8h TWA) | Action Level |
|----------|-------------------------------------|--------------|
| OSHA PEL (USA) | 90 dB(A) | 85 dB(A) |
| NIOSH REL (USA) | 85 dB(A) | — |
| EU Directive 2003/10/EC | 87 dB(A) | 80 dB(A) (lower) / 85 dB(A) (upper) |
| ISO 1999:2013 | 85 dB(A) (recommended) | — |

**Halving/doubling rule:** For every 3 dB increase (NIOSH/EU) or 5 dB increase (OSHA), permissible exposure time halves:
- NIOSH: 85 dB → 8h, 88 dB → 4h, 91 dB → 2h, 94 dB → 1h
- OSHA: 90 dB → 8h, 95 dB → 4h, 100 dB → 2h, 105 dB → 1h

### Environmental Noise Limits (Typical)

These vary by jurisdiction. Common residential limits:

| Time Period | Typical Limit | Standard/Regulation |
|------------|---------------|---------------------|
| Daytime (7am–10pm) | 55 dB(A) Leq | WHO guidelines, many local codes |
| Nighttime (10pm–7am) | 45 dB(A) Leq | WHO guidelines |
| Lmax nighttime | 60 dB(A) | WHO sleep disturbance threshold |

Always ask the user which specific regulation they're working to — limits vary significantly by location and zone (residential, commercial, industrial).

### Peak Level Interpretation

| Lpeak dB(C) | Risk |
|-------------|------|
| < 135 | Generally safe for single events |
| 135–140 | EU upper exposure action value — hearing protection mandatory |
| > 140 | Immediate acoustic trauma risk |

### Output Metric Selection

| User's Question | Which Output to Report |
|-----------------|----------------------|
| "What's the noise level?" | Leq (time-averaged level — single representative value) |
| "How loud does it get?" | Lmax (maximum level during measurement) |
| "Is there an impulse hazard?" | Lpeak (true peak pressure, unweighted by time constant) |
| "What does the level look like over time?" | Lt (time-weighted instantaneous level — use downsampled intervals) |
| "8-hour exposure?" | Leq over measurement period, then compute TWA |

## Common Scenarios

### Scenario: Background Noise Survey for NC Rating

1. Record ≥15 s of steady-state background noise (HVAC on, room unoccupied)
2. Compute A-weighted Leq as a quick sanity check:
   ```
   splMeter: Bandwidth="Full band", FrequencyWeighting="A-weighting"
   ```
   Compare to the A-weighted target for the space type (see `references/noise-compliance.md`)
3. Compute octave-band Leq for spectral diagnosis:
   ```
   splMeter: Bandwidth="1 octave", FrequencyWeighting="Z-weighting", FrequencyRange=[15 8000]
   ```
4. For formal NC/RNC/RC compliance, pass the recording to `noiseCriteria(audioIn, fs, CalibrationFactor=cf)` — it computes octave-band Leq and assigns the rating internally

### Scenario: Product Noise Measurement

1. Set up mic at specified distance (typically 1 m, per product standard)
2. Record product in specified operating conditions
3. Compute A-weighted Leq: `splMeter` with `Bandwidth="Full band"`, `FrequencyWeighting="A-weighting"`
4. Report `Leq(end)` as the product noise level in dB(A)
5. If spec requires octave-band data, also compute with `Bandwidth="1 octave"`, `FrequencyWeighting="Z-weighting"`

### Scenario: Occupational Noise Assessment

1. Record representative work period (or use duration-weighted samples)
2. Compute A-weighted Leq with Slow time weighting:
   ```
   splMeter: Bandwidth="Full band", FrequencyWeighting="A-weighting", TimeWeighting="slow"
   ```
3. Report Leq(end) and compare to exposure limit
4. If peak hazard is a concern, also report Lpeak with C-weighting:
   ```
   splMeter: FrequencyWeighting="C-weighting"
   ```

## Troubleshooting

### Handling splMeter output size

`splMeter` outputs one row per input sample, but `Leq` is only updated at `TimeInterval` boundaries — between boundaries the value is held from the previous interval, and it is `-Inf` before the first boundary. Choose `TimeInterval` based on what you need:

| Goal | TimeInterval | How to read Leq |
|------|-------------|-----------------|
| Single overall level | Set to signal duration | `Leq(end)` |
| Time history (e.g., 1-second snapshots) | `1` | `Leq(fs:fs:end,:)` (value at each 1-second boundary) |

**Notes:**
- If `TimeInterval` > signal duration, `Leq` is never reported (all `-Inf`)
- `Leq` resets at each `TimeInterval` boundary — it is NOT a cumulative running total across the full signal (see "TimeInterval and Leq accumulation" above)

### Common issues

| Observation | Likely Cause | Action |
|-------------|-------------|--------|
| SPL seems 20–40 dB too low or too high | Missing or incorrect calibration factor | Verify CalibrationFactor; if unknown, levels are relative only |
| Leq is all -Inf | TimeInterval > signal duration | Set TimeInterval ≤ signal duration |
| Octave-band output missing expected bands | FrequencyRange excludes some true center frequencies | Use `[15 8000]` for 16 Hz–8 kHz; see SKILL.md gotchas |
| All octave bands show same level | FrequencyWeighting is applied — Z-weighting for flat bands | Use Z-weighting if you want unweighted per-band levels |
| Levels don't match physical meter | Different weighting, time constant, or calibration | Match all settings: same weighting, same time constant, verify cal |
| Lpeak is much higher than Leq | Normal for impulsive or highly variable noise | Report both; Lpeak represents instantaneous peak exposure |

----

Copyright 2026 The MathWorks, Inc.

----
slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."Stars: 198kUpdated Claude CodeCodexVerifiedInstall
ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt.Stars: 194kUpdated Claude CodeCodexVerifiedInstall