Dugan方式ゲインシェアリングで複数人のポッドキャスト音声を自動ミキシングする方法
複数マイクでのかぶり（ブリード）と位相干渉を解消：サブサンプル精度の位相整合、Dugan方式の自動ゲイン配分、LUFS規格準拠。
執筆： Divy YadavYouTuber／AutoSilenceCut 開発者目次 (6)↓
The Multi-Microphone Podcast Dilemma: Bleed & Phase Lag
When recording a round-table podcast or multi-person interview with 2 to 4 hosts, each speaker's voice bleeds into the adjacent microphones. Because sound travels through air at approximately 343 meters per second, the voice hits the secondary microphone 5 to 25 milliseconds later.
Summing these delayed signals together creates comb filtering and background ambient buildup, making the podcast sound like it was recorded in an echoey cave.
How Dugan-Style Gain-Sharing Automixing Eliminates Cross-Talk
Invented by audio pioneer Dan Dugan, speech automixing distributes total system gain across all active channels based on relative speech energy:
When the host speaks, their channel receives 100% gain, while the guest microphone is automatically attenuated by 12dB to 18dB. When both speakers interrupt each other, gain is apportioned dynamically so total acoustic power remains constant at 0dB without digital clipping.
Sub-Sample Phase Delay Alignment via Pearson Correlation
AutoSilenceCut calculates Pearson normalized cross-correlation across stems over a 50ms search window. By finding the exact delay index that maximizes correlation, the engine shifts secondary track samples forward or backward, locking waveforms into phase coherence.
Achieving -16 LUFS Integrated Loudness for Apple & Spotify
Major podcast distribution platforms enforce strict broadcast loudness targets (-16 LUFS for stereo, -19 LUFS for mono). AutoSilenceCut normalizes the balanced master mix using ITU-R BS.1770-4 K-weighting with an inter-sample True-Peak limiter ceiling of -1.0 dBFS.
Step-by-Step Multi-Stem Automixing Workflow
- Open the Studio: Navigate to the Multi-Track Podcast Automixer Studio.
- Load Stems: Upload separate audio files for Host, Co-Host, and Guest microphones.
- Verify Alignment: Review detected phase lag and Dugan automix weighting.
- Audition Mix: Play the balanced mix and test individual stem Mute/Solo channels.
- Export Master: Download the aligned broadcast WAV master directly.
この手順を次のツールで実践： Multi-Track Podcast Automixer Studio
Web Audio APIによる100%ローカル処理。音声の外部送信ゼロ、非圧縮WAV出力。
技術的な質問と解説
Why does microphone bleed ruin multi-person podcast mixes?
When multiple microphones are open in the same room, each mic picks up the other speaker with a small acoustic delay. When summed together, this creates phase cancellation and hollow reverberant bleed.
How does Dugan-style gain-sharing automixing differ from standard noise gates?
Traditional noise gates abruptly mute and unmute audio, causing distracting pumping and clipped word beginnings. Dugan automixing continuously shares total gain mathematically, smoothly dipping inactive mics without harsh gating.
Is multi-track automixing private in AutoSilenceCut?
Yes. All stem decoding, Pearson cross-correlation, Dugan gain calculations, and WAV rendering execute 100% locally in your browser memory.
主要参考文献・音響規格基準
- Dan Dugan (1974) — Automatic Microphone Mixer, Journal of the Audio Engineering Society: Foundational mathematical specification for automatic speech level gain sharing and acoustic comb-filter prevention in multi-microphone dialogue systems.[link]
- ITU-R BS.1770-4 — Algorithms to Measure Audio Programme Loudness and True-Peak Audio Level: International Telecommunication Union global standard defining K-weighted integrated loudness (LUFS/LKFS) and 4x oversampled true-peak measurement.[link]
会話音声と音響の補足規格
SRT・WebVTT字幕ファイルを読み込んで音声編集箇所を目視チェックする方法
テキストと音声の同期システム：字幕ファイルの取り込み、文章内の意図的な間の保護、タイムコード再計算による同期字幕の書き出し。
声の編集に最適な無料オーディオエディタの選び方とローカル処理の重要性
会話音声向けツールの5大評価基準：ブラウザ内ローカル処理によるプライバシー保護、非破壊可逆編集、会話テンポ制御、非圧縮WAV出力。
Webブラウザ上で完結するオンライン音声編集：安全で可逆的な整音手法
Web Audio APIによるクライアントサイド処理の仕組み：サーバーへのアップロード不要、メモリ上でのプレビュー検証、オリジナル音源の保持。