audio_defect_detector 0.0.1

Pure Dart library to detect pops and clicks in WAV audio files using adaptive threshold analysis.

dart_pop_n_click

Pure Dart library to detect pops and clicks in WAV audio files.

Features

• Detect clicks (1–10 samples) and pops (10–150 samples) in WAV audio
• Adaptive MAD-based threshold with configurable sensitivity
• Support for PCM 8/16/24/32-bit and IEEE Float 32-bit WAV files
• Raw PCM byte analysis without WAV headers
• Pre-normalised sample analysis for integration with other decoders
• Per-channel or mono-summed analysis
• Confidence scoring with logistic function
• CLI tool with text and JSON output, glob support, and batch processing
• Public median() and mad() statistical utilities for custom signal analysis
• Pure Dart — no native dependencies, works on all platforms

Installation

dart pub add audio_defect_detector

Package: audio_defect_detector

A pure-Dart, cross-platform library for detecting transient audio defects (pops and clicks) in WAV files, with a companion CLI tool.

---

Library API

import 'package:audio_defect_detector/audio_defect_detector.dart';

void main() async {
  final result = await analyseFile(
    'recording.wav',
    config: DetectorConfig(
      sensitivity: Sensitivity.medium,
      minConfidence: 0.5,
    ),
  );

  print('Aggregate confidence: ${result.aggregateConfidence}');
  for (final d in result.defects) {
    print('${d.offset} – ${d.type.name} (confidence: ${d.confidence.toStringAsFixed(3)})');
  }
}

Public API Summary

Module	Exports
analyser.dart	analyseFile(), analyseBytes(), analysePcm(), analyseSamples()
models.dart	DetectorConfig, Sensitivity, Defect, DefectType, AudioMetadata, AnalysisResult, PcmFormat, UnsupportedFormatException, CorruptFileException, IoException
pcm_decoder.dart	decodePcmBytes()
math_utils.dart	median(), mad()

analyseFile(path, {config})

Reads the file at path and returns a Future<AnalysisResult>.

analyseBytes(bytes, {path, config})

Analyses a Uint8List of audio bytes and returns a Future<AnalysisResult>.

DetectorConfig

Field	Type	Default	Description
sensitivity	Sensitivity	medium	Controls threshold aggressiveness (low, medium, high).
minConfidence	double	0.0	Suppress results below this score.
maxDefects	int	0	Limit total results (0 = unlimited).
perChannel	bool	false	Analyse channels independently.

AnalysisResult

Field	Description
defects	List<Defect> sorted by time offset.
aggregateConfidence	Overall file-level defect probability (0.0–1.0).
metadata	AudioMetadata (sample rate, bit depth, channels, duration).

Defect

Field	Description
offset	Duration from the start of the file.
length	Duration of the defect span.
type	DefectType.click (1–10 samples) or DefectType.pop (10–150 samples).
confidence	Score 0.0–1.0 derived from peak-to-noise ratio via logistic function.
channel	Zero-based channel index.
sampleIndex	Sample index of the peak anomaly.
amplitude	Normalised peak amplitude (–1.0 to 1.0).

Exceptions

• UnsupportedFormatException — unsupported file format.
• CorruptFileException — malformed or truncated file.
• IoException — I/O error (file not found, permission denied).

---

Raw PCM Analysis

For audio that has already been decoded to raw PCM bytes (e.g. from FLAC decoders):

import 'package:audio_defect_detector/audio_defect_detector.dart';

// Analyse raw 16-bit signed LE stereo PCM bytes
final result = await analysePcm(
  pcmBytes,
  format: PcmFormat(sampleRate: 44100, bitDepth: 16, channels: 2),
);

For pre-normalised Float32 samples:

// samples is List<Float32List> — one per channel, values in [-1.0, 1.0]
final result = await analyseSamples(
  samples,
  sampleRate: 44100,
);

---

Statistical Utilities

The library exposes two robust statistical functions from math_utils.dart for custom signal analysis:

median(Float32List sorted)

Compute the median of a sorted Float32List.

mad(Float32List values)

Median Absolute Deviation — a robust dispersion measure that is resistant to outliers.

import 'package:audio_defect_detector/audio_defect_detector.dart';

final values = Float32List.fromList([1.0, 2.0, 3.0, 4.0, 100.0]);
final dispersion = mad(values); // 1.0 — robust to the outlier

---

CLI: audiodefect

Activate globally:

dart pub global activate audio_defect_detector

Usage

audiodefect analyse <file|glob> [options]

Options

Flag	Description
--sensitivity=low|medium|high	Detection sensitivity (default: medium).
--min-confidence=0.0–1.0	Suppress results below this score.
--threshold=0.0–1.0	Exit code 1 when defects above this score are found.
--output=text|json	Output format (default: text).
--quiet	Suppress all output.
--verbose	Show extra diagnostics.
--raw	Treat input as raw PCM (no WAV header).
--sample-rate=N	Sample rate for raw PCM (default: 44100).
--bit-depth=N	Bit depth for raw PCM: 8, 16, 24, or 32 (default: 16).
--channels=N	Number of channels for raw PCM (default: 2).
--float	Treat raw PCM as IEEE float instead of integer.

Exit codes

Code	Meaning
0	Clean — no defects above threshold.
1	Defects found above threshold.
2	Usage error.
3	File / decode error.

Example — text output

$ audiodefect analyse recording.wav

File : recording.wav
Audio: 2ch  44100 Hz  16-bit  0m03.000s
Score: aggregate confidence = 0.997

2 defect(s) found:

 Offset (ms)    Type    Conf  Channel   Amplitude
────────────────────────────────────────────────────────
        1000   click   1.000        0      0.9900
        2000     pop   1.000        0      0.9000

Example — JSON output

$ audiodefect analyse recording.wav --output=json
{
  "file": "recording.wav",
  "result": {
    "schema_version": "1",
    "aggregate_confidence": 0.997,
    "defect_count": 2,
    "metadata": { "sample_rate": 44100, "bit_depth": 16, "channels": 2, "duration_ms": 3000 },
    "defects": [...]
  }
}

Example — raw PCM analysis

$ audiodefect analyse recording.raw --raw --sample-rate=48000 --bit-depth=24 --channels=1

---

Detection Algorithm

1. High-pass differentiator: d[n] = x[n] - x[n-1]. Accentuates transients.
2. Adaptive MAD threshold: Median Absolute Deviation over a sliding ≈10 ms window provides a local noise estimate. The threshold is MAD × 1.4826 × thresholdMultiplier.
3. Region merging: Adjacent flagged samples within a gap of 4 are merged into contiguous regions.
4. Classification: Regions 1–10 samples wide → click; 10–150 samples → pop; wider → discarded.
5. Confidence score: Logistic function applied to peak-to-noise ratio, yielding a calibrated 0.0–1.0 score.

---

Supported Formats

Phase	Format	Status
1	WAV PCM 8/16/24/32-bit, mono/stereo	✅ Supported
1	WAV IEEE Float 32-bit	✅ Supported
2	FLAC 16/24-bit, mono/stereo	🔜 Planned

---

Running Tests

dart pub get
dart test

---

Limitations

• Maximum file size: 2 GB (files are loaded entirely into memory)
• Only WAV format is currently supported (FLAC planned for future release)
• Glob support in CLI is limited to simple prefix*suffix patterns