Transparency and data sovereignty are at the foundation of ClearVoice. Whether you are an investigative journalist, commercial podcast producer, educator, or privacy-conscious creator, you have a right to know exactly what happens to your audio and video recordings when they are processed in the cloud. This disclosure outlines our neural speech isolation architecture, broadcast vocal mastering specifications, and automated zero-retention cloud deletion policy.
1. AI Neural Speech Isolation Engine
ClearVoice runs on its own AI Audio Engine, a deep neural network architecture designed for full-band speech enhancement. Unlike older spectral subtraction algorithms that create unnatural "underwater" musical noise or harsh robotic artifacts, our engine processes audio through complex spectral filtering across 512 discrete frequency bins:
- Acoustic Rumble Cut: An initial 80Hz high-pass filter cuts sub-bass HVAC rumble, wind buffeting, and microphone handling vibrations.
- Non-Stationary Noise Isolation: Suppresses sudden transient spikes such as honking cars, slamming doors, and keyboard typing.
- AI Consonant Guard: Protects low-energy vocal transients (plosives, sibilants, and whispered word endings) from being clipped or muffled.
- Adjustable AI Strength: Users can select between 50% Natural (gentle room tone), 70% Studio (clean dialogue), or 100% Maximum AI (complete isolation).
2. Studio Vocal Mastering Chain (-16 LUFS)
Clean dialogue alone is not enough for professional release; vocals recorded on laptops, iPhones, or lavalier mics often sound thin and unbalanced. Directly following neural noise isolation, ClearVoice routes speech through a hardware-modeled mastering chain:
- 220Hz Chest Resonance Warmth (+1.5 dB): Restores body and authority to thin recordings.
- 3.2kHz Vocal Presence (+3.2 dB): Modeled after the frequency response of the industry-standard Shure SM7B broadcast dynamic microphone to ensure speech cuts through any background music.
- 8kHz Acoustic Brilliance (+1.2 dB): High-shelf polish adding acoustic air and modern studio clarity.
- Studio Dynamics Compressor: A leveling compressor (threshold: -18dB, ratio: 2.5:1, attack: 20ms, release: 140ms) balances whisper-to-shout dynamics into smooth, broadcast-ready dialogue.
- EBU R128 Loudness Normalization (-16 LUFS): Delivers standardized loudness compliant with YouTube, Spotify, and Apple Podcasts.
3. 100% Video Frame Integrity (Zero Re-encoding)
When you upload MP4, MOV, or MKV videos, ClearVoice never touches your video track. The video frames are demuxed, the audio is extracted and enhanced, and the pristine audio is remultiplexed with ffmpeg -c:v copy:
- Zero Quality Loss: 4K, 1080p, 60fps, and high-bitrate video streams remain 100% byte-for-byte identical to your original upload.
- Zero Color Shift: Color profiles, HDR metadata, and container flags are completely preserved.
4. Automated Zero-Retention Cloud Storage Lifecycle
ClearVoice enforces strict, automated cloud storage deletion on all customer media via Cloudflare R2:
| Scenario | Retention Window | Action |
|---|---|---|
| Downloaded Files | 15 Minutes | Automatically and permanently purged from cloud storage 15 minutes after download. |
| Un-Downloaded Files | 2 Hours | Automatically destroyed after 2 hours if not downloaded or claimed. |
| On-Demand Wipe | Immediate (0s) | Clicking "Clean another file" purges the previous file and intermediate stems instantly. |
| AI Training | Never | Your files are never stored permanently, shared, or used to train public machine-learning models. |