Welcome to the official documentation for the Sonodit Audio Pipeline. Our platform autonomously transforms raw voiceover recordings into broadcast-ready, mastered assets. Below, we detail the chronological journey your audio takes within our intelligent processing engine:

Step 1: Ingestion and Acoustic Profiling

Upon file upload, the Sonodit Audio Profiler performs a comprehensive signal analysis. We measure sample rate, bit depth, and initial crest factor. Simultaneously, a neural network maps the speaker’s phonetic structure to determine precise Words Per Minute (WPM), native language, and regional dialect. This generates a predictive roadmap that guides all subsequent processing tools.

Step 2: Subsonic Filtering and Noise Subtraction

To ensure absolute purity, the audio enters the restoration phase. A dynamic High-Pass Filter set at 50 Hz is applied to eliminate parasitic low-end rumble. Concurrently, the engine identifies the 'noise fingerprint' detected during speaker pauses and applies a 50% inverse spectral subtraction algorithm, effectively silencing room hiss without compromising the natural timbre of the voice.

Step 3: Block-Based Stabilisation (Clip Gain Matching)

One of our most sophisticated stages: the audio undergoes dual-stage normalisation. First, the entire signal is standardised to -25.0 dB RMS. Immediately following, the system analyses the audio phrase-by-phrase. Whether the speaker whispered a line or delivered a punchy slogan, the algorithm calculates exact volume variances and applies independent corrections with a +/- 15.0 dB safety limiter, preserving short pauses through a precision exclusion window. This ensures every word maintains consistent presence and commercial impact.

Step 4: Dynamic Isolation (Predictive Gating)

The stabilised audio passes through an ultra-precise noise gate configured between -50 dBFS and -55 dBFS. The system detects voice attack transients to open the signal path instantly, applying controlled fades on word tails to bury any unwanted breath or residual air in absolute silence.

Step 5: Acoustic Modelling and Commercial Texturing

The cleaned file is injected into our virtualised DSP engine. Here, the audio is processed through a simulated analogue rack that applies soft-knee compression, surgical de-essing, and parallel harmonic excitation. This phase amalgamates the signal, bestowing the voice with that classic 'Fat and Clear' sound characteristic of premium radio networks and cinematic productions, with a strict True Peak limit of -2.0 dB.

Step 6: Precision Segmentation and PTK Shielding

In the final stage, our temporal orchestrator (The Cartographer) assumes control. For autonomous workflows, it identifies silence valleys and performs precise cuts to export phrases as independent files. Each segment includes a 10-millisecond PTK (Pre-Trigger Kill) protection margin, ensuring no initial consonants are clipped. If a unified master is requested, the system assembles the phrases by injecting dynamic silences calculated from the speaker's natural rhythm, resulting in a compact, rhythmic piece ready for immediate delivery.