ノーマライズ（音量均一化）vs コンプレッサー：音声における違いと使い分け

Core Mechanics: Static Gain vs Dynamic Modulation

Among new podcasters and spoken-word creators, "normalizing" and "compressing" are frequently conflated as interchangeable volume tools. In reality, they operate on completely different mathematical and acoustic principles.

How Audio Normalization Works (Linear Gain)

Normalization is a static, linear mathematical operation. The software scans an audio file, calculates a single global gain offset in decibels, and applies that exact same volume shift to every single audio sample from start to finish.

• Linear Relationship: If sample A is 6 dB louder than sample B before normalization, sample A will remain exactly 6 dB louder than sample B after normalization.
• No Dynamic Modulation: Normalization does not compress peaks, boost quiet syllables, or alter the crest factor of the recording.

How Audio Compression Works (Dynamic Envelope)

Compression is a non-linear, dynamic volume automation process. An audio compressor continuously monitors the incoming signal level relative to a user-defined threshold. When the volume crosses that threshold, the compressor temporarily reduces the volume by a specified ratio.

• Dynamic Range Reduction: By squashing loud transient spikes (e.g., sudden laughter or enthusiastic emphasis), the compressor narrows the distance between the quietest and loudest parts of the voice.
• Makeup Gain: Once the peaks are controlled, makeup gain lifts the entire track, making quiet phrases much easier to understand in noisy environments (like cars or gym earbuds).

Peak Normalization vs Loudness (LUFS) Normalization

Not all normalization methods are created equal. Understanding the difference between Peak and LUFS normalization is essential for modern podcasting.

The Flaw of Peak Normalization: The Single Laugh Trap

Peak Normalization scans for the single highest sample voltage in the file (e.g., -4.0 dBFS) and boosts the entire track so that peak hits 0.0 dBFS (a +4.0 dB boost).

The Fatal Flaw: If a podcast guest coughs or laughs loudly at minute 14 hitting -0.5 dBFS, but speaks quietly at -28 dBFS throughout the rest of the episode, Peak Normalization will only boost the file by a negligible +0.5 dB. The dialogue remains quiet, muddy, and unintelligible.

Loudness Normalization: Human Hearing & ITU-R BS.1770

Loudness Normalization (LUFS) uses psychoacoustic filtering (K-weighting curves defined by ITU-R BS.1770-4) that mimics human ear frequency sensitivity (attenuating sub-bass and boosting 2kHz–4kHz presence).

It measures the integrated perceived loudness of the dialogue over time, ignoring short transient peaks. This ensures your podcast plays at the exact same perceived volume as professional shows on Spotify, Apple Podcasts, and YouTube.

Comparison Matrix: Normalization vs Compression

Evaluate these technical distinctions when mastering in the Audio Normalizer:

Signal Chain Order: When to Compress and When to Normalize

To achieve professional broadcast clarity, follow this strict signal sequence:

1. Subtractive Cleanup: Apply an 80Hz high-pass filter and 50/60Hz mains hum notch filter. Removing inaudible sub-bass prevents compressors from false-triggering on desk bumps.
2. Silence & Transient Editing: Use AutoSilenceCut to prune awkward dead air (with 0.32s speech cushions) and de-click mouth sounds.
3. Dynamic Vocal Compression: Apply gentle compression (Ratio: 2.5:1, Attack: 30ms, Release: 150ms) to tame erratic volume spikes by 2–4 dB.
4. True-Peak Safety Limiting: Set a brickwall limiter ceiling at -1.0 dBTP to catch inter-sample overshoots.
5. Final ITU-R BS.1770 Normalization: Scale the finished audio to -16.0 LUFS Integrated for stereo distribution.

実践ケーススタディ：The Quiet/Loud Interview Guest

An audio engineer receives an interview track where the host speaks consistently at -18 LUFS, but the remote guest alternates between whispering intimate anecdotes (-28 LUFS) and bursting into loud laughter (-8 LUFS peak).

• Why Normalization Alone Fails: Normalizing the guest's track directly to -16 LUFS would boost the quiet whispers by +12 dB, but the loud laughter would smash into digital clipping distortion.
• The Correct Hybrid Solution:
  1. Apply a 3:1 compressor to reduce the loud laughter peaks by 5 dB.
  2. Apply gentle upward makeup gain (+4 dB) so whispers become clear.
  3. Run final LUFS Loudness Normalization to lock the master file to -16.0 LUFS.
• Result: Smooth, effortless listening where the audience never has to reach for the volume knob.

音声編集・処理で避けるべき代表的な失敗と注意点

• Mistake 1: Relying on Peak Normalization for Delivery: Peak normalization is blind to human loudness perception. Always master to Integrated LUFS.
• Mistake 2: Over-Compressing Dialogue (The "Squashed Radio" Effect): Setting a high ratio (8:1) with fast attack crushes vocal dynamics, turns breath sounds into loud gasps, and makes speech exhausting to hear.
• Mistake 3: Normalizing Before Cleaning Noise: If you normalize a quiet recording with background hiss, you boost the hiss along with the voice. Clean noise first, normalize last.
• Mistake 4: Forgetting the -1.0 dBTP True Peak Margin: Exporting at 0.0 dBFS guarantees distortion when MP3 encoders convert PCM waveforms into lossy codecs.