Beamie Voice Synthesis & Audio DSP Pipeline

Overview

Beamie is CargoBeamer’s official corporate AI mascot, created to communicate digital transformation, sustainability metrics, and operations updates across the company and European logistics corridors.

To give Beamie a consistent, high-fidelity acoustic voice across English, German, French, and Italian, I engineered an automated CLI and audio DSP pipeline that takes raw speech synthesis and transforms it into broadcast-grade master audio.

Engineering Highlights

1. Multi-Model Synthesis Engine:

   • Designed a modular synthesis backend supporting high-quality cloud models (ElevenLabs Bella Multilingual v2) alongside local, low-latency offline fallback models (Kokoro-FastAPI, Edge-TTS).
   • Standardized phoneme mapping, pronunciation overrides for proprietary rail freight terminology (e.g., intermodal, pocket wagon, horizontal loading), and multi-lingual cadence matching.

2. 5-Stage Studio Audio DSP Chain:

   • Stage 1 (Pre-EQ Filtering): Applied 85 Hz highpass filtering to eliminate sub-bass rumble, coupled with a surgical 320 Hz de-mud notch (-2.5 dB, Q=1.8).
   • Stage 2 (Presence & Clarity): Tuned a 3.6 kHz bell boost (+3.0 dB, Q=1.2) for vocal intelligibility across phone speakers and conference room acoustics.
   • Stage 3 (High-Frequency Air): Introduced a 10.5 kHz high-shelf (+2.0 dB) for modern broadcast sheen.
   • Stage 4 (Haas Stereo Widening): Implemented micro-delay stereo widening (55 ms delay, 15% wet mix) to create subtle spatial separation without phase cancellation in mono playback.
   • Stage 5 (Dynamics & Loudness Normalization): Configured optical compression (2.5:1 ratio, 30 ms attack, 120 ms release) and strict EBU R128 loudness normalization (-16.0 LUFS integrated loudness, -1.5 dBFS true peak).

3. Developer CLI & Video Pipeline Integration:

   • Packaged the entire DSP engine into a streamlined Python CLI tool (beamie).
   • Integrated with Remotion video pipelines to automatically align generated speech waveforms with dynamic animated lip-sync and timed on-screen captions.

Business Impact

• Consistent Brand Identity: Established a unified acoustic fingerprint for CargoBeamer across all multimedia collateral and product tutorials.
• Fast Production Turnaround: Reduced voice-over production from days (waiting on external voice talent) to seconds via automated CLI scripts.
• Enterprise Asset Governance: Maintained strict canonical source-of-truth guidelines and audio master assets across company repositories.