ナレッジハブ・技術ガイド
音声ノイズ除去・ノーマライズおよび音質向上
音声ノイズ除去・ノーマライズおよび音質向上
本ツールは100%クライアントサイド（ブラウザ内）で動作します。写真、動画、音声、PDFファイルが外部サーバーにアップロードされることは一切なく、お使いの端末のメモリ内で安全に処理されます。
コアアーキテクチャ & 計算式
Clean Audio = Raw Signal - Noise Profile (FFT Thresholding)
The algorithm analyzes a 'silent' portion of the audio to build a 'Noise Profile'. It then scans the entire file, acting as a microscopic volume knob (spectral gate) to turn down the specific frequencies that match the noise.
ベストプラクティスとガイドライン
- Always Record Room Tone: The algorithm needs to know what the 'noise' sounds like. Always record 10 seconds of absolute silence in your room before you start speaking. This provides the perfect Noise Profile for the cleaner.
- Target the 50/60Hz Hum: Electrical interference (like a bad microphone cable) creates a constant low frequency hum. Use a highly targeted notch filter exactly at 50Hz (Europe) or 60Hz (US) to surgically remove it.
- Don't Over Process: If you set the noise reduction threshold too high, the algorithm will start eating into the vocal frequencies, making the speaker sound robotic or like they are talking underwater (artifacting).
よくある質問 (FAQ)
Can this remove a dog barking in the background?
Generally no. Spectral gating works best on constant, continuous noise (like an air conditioner or fan hiss). Sudden, loud, dynamic noises (like a dog bark or a door slam) occupy the same frequencies as human speech and are incredibly difficult to remove.
Does this tool fix clipped audio?
No. If your microphone was too loud and the audio 'clipped' (hit the red zone and distorted), the waveform is permanently truncated. No software can magically invent the audio data that was destroyed.
Why did my audio get quieter?
Noise often adds artificial volume to a track. When you strip the noise away, the perceived volume drops. You will usually need to apply 'Normalization' or a slight gain boost after cleaning.