オーディオエンハンサー vs ノイズ除去：それぞれの得意・不得意

Subtractive Cleanup vs Additive Enhancement: Core Principles

In modern audio production, confusion between "enhancing" and "denoising" causes more ruined recordings than any other misconception. Understanding the mathematical difference between subtractive DSP and additive DSP is essential for professional spoken-word delivery:

• Subtractive Audio Restoration (Noise Removers): Isolates unwanted acoustic energy that does not belong to the human voice and removes or attenuates it. This includes high-pass filtering below 80 Hz, notch filtering 50/60 Hz electrical hum, and downward expansion during pauses. Subtractive tools reduce total signal energy to reveal the clean voice underneath.
• Additive Audio Enhancement (Vocal Enhancers): Takes a clean voice signal and enhances its acoustic character. This includes shaping vocal presence (boosting clarity at 3.5 kHz so consonants cut through), adding high-frequency sheen (10 kHz air), warming fundamental chest resonance (200 Hz), and applying gentle soft-knee compression to even out volume fluctuations.

The Cardinal Rule: Why Enhancers Amplify Noise on Raw Tracks

Audio enhancers use presence equalization, harmonic saturation, and upward dynamic makeup gain. When you feed a noisy recording directly into an enhancer:

1. The 3.5 kHz presence boost amplifies computer fan hiss and room reflection flutter.
2. The 10 kHz air shelf boosts microphone preamp thermal self-noise into an irritating whistle.
3. The dynamic makeup compression pulls quiet background noise up in volume every time the speaker takes a breath or pauses.

The result is a track where the background noise pumps aggressively and competes with the speaker’s voice.

Defect Diagnosis Matrix: Enhancer vs Noise Remover

Consult this diagnostic matrix to match your specific audio defect with the correct DSP tool:

The Correct Dialogue Signal Chain: Subtractive Before Additive

To achieve pristine broadcast-grade voiceover or podcast audio, route your signal through this non-destructive 4-step hierarchy:

ステップ 1：Subtractive Foundation (High-Pass, Notch, Expander)

First, eliminate unwanted non-speech frequencies. Cut below 80 Hz with a 12 dB/octave high-pass filter to remove subsonic rumble, engage surgical 50/60 Hz notch filtering for ground loops, and use a downward expander (-38 dBFS threshold) to smoothly lower the noise floor during pauses.

ステップ 2：Corrective Resonance Carving (Boxiness Removal)

Untreated rooms create resonant acoustic peaks between 300 Hz and 600 Hz (often called "cardboard box" or "nasal" resonance). Use a narrow parametric EQ cut (-2 dB to -3 dB at 450 Hz with a Q of 2.5) to clear room congestion before boosting any vocal frequencies.

ステップ 3：Additive Vocal Presence & Air Sheen EQ

Now that the noise floor is pristine and resonances are carved out, apply additive enhancement:

• Vocal Presence (3.5 kHz, Broad Bell): Boost by +2 dB to +3.5 dB. This enhances consonant intelligibility ("t", "d", "c", "p") and brings the speaker forward in the mix.
• Air Sheen (10 kHz, High-Shelf): Boost by +1.5 dB to +2.5 dB. This provides modern broadcast sparkle and breath clarity without harshness.
• Vocal Warmth (200 Hz, Low-Mid Bell): Boost by +1 dB to +2 dB for thin voices to restore chest resonance and intimate authority.

ステップ 4：True-Peak Headroom Protection (-1.0 dBFS)

Finish with a brickwall true-peak limiter set with a ceiling of -1.0 dBFS. This transparently catches sudden plosive peaks or excited vocal shouts, preventing inter-sample digital clipping during MP3 compression and streaming distribution.

Studio Case Study: Transforming a Muffled USB Mic Recording

Case Study: An online educator recorded a 45-minute technical course on a $60 USB dynamic microphone in a home office. The voice sounds dark, muffled, and distant, with a steady 60 Hz electrical hum from the laptop power supply and a continuous -43 dBFS room noise floor.

The Combined Cleanup & Enhancement Pipeline:

1. Subtractive Noise Cleanup: Applied an 80 Hz high-pass filter and a 60 Hz narrow notch filter ($Q=10$). Low-end hum and rumble dropped by -22 dB.
2. Downward Expansion: Set threshold to -40 dBFS with a 1:2 ratio. The quiet room hiss dropped from -43 dBFS to -53 dBFS during sentence breaks.
3. Vocal Presence Boost: Applied +3.5 dB presence boost at 3.2 kHz. The muffled, blanket-covered character disappeared, revealing crisp, articulated speech consonants.
4. Air Sheen: Added +2 dB high shelf at 10 kHz to impart an open, high-end studio feel.
5. Loudness Normalization: Normalized output to -16 LUFS Integrated (YouTube/Podcast standard) with a -1.0 dBFS true-peak ceiling.

The "Do Not Process" Decision Gate: When to Reject & Re-Record

Software tools have strict physical boundaries. If a recording exhibits any of the following four conditions, DSP enhancement will only make the audio sound more artificial. Stop editing and re-record at the source:

• Severe Digital Clipping (Flat-Topped Waveforms): If the preamp was overloaded and speech waveforms are squared off at 0 dBFS across entire sentences, the harmonic data is permanently destroyed.
• Heavy Slapback Echo (RT60 > 0.8s): If the speaker recorded in a large tiled bathroom or empty glass conference room, the reverberant energy equals the direct speech energy. De-reverberation will render the voice robotic.
• Excessive Distance (Over 24 Inches from Capsule): When the mic is too far, room noise and reflections overpower vocal fundamentals. Boosting presence only magnifies the room flutter.
• Overlapping Cross-Talk: If two hosts spoke simultaneously on a single track, DSP filters cannot separate the overlapping voice formants without severe phase cancellation.