Intelligent Dynamics

Different language.
Same feeling.

Examples

AUTONOMOUS TRANSLATION / VOICE STUDIES

Joseph Stalin, in English.

READY
0:000:12
Ready to listenGenerated WAV · English

Synthetic speech from prepared text. Compare with the paired archival excerpt.

INTELLIGENT DYNAMICS

The audio side of Intelligent Dynamics.

Player build: librelm-page-05

This listening room is our audio showcase: recorded speech translation, expressive dubbing and text-to-speech. It is one part of our broader work on robust, purpose-built models for data-constrained environments.

DATA EFFICIENCY / PURPOSE-BUILT INTELLIGENCE

We design custom training pipelines around the problem and the data available. Our fine-tuning pipeline and Jeju TTS system are implemented; achieving the same quality with fewer task-specific recordings remains a goal to validate.

BEYOND AUDIO

Our other work includes an extremism classifier targeting lower-cost, task-specific inference, and LibreLM, our exploration of compact language-model training and pipeline reuse.

The founders’ research and engineering also spans a hybrid Mamba–Transformer masked-diffusion model, language models and retrieval, and agentic machine learning for high-dimensional biomedical data.

Online slideshow Static PDF Original PowerPoint

About these recordings
Autonomous translation
Our workflow is designed to translate and dub speech autonomously. The historical examples here are six recorded voice studies made from prepared target-language text, not a live or validated end-to-end translation test. Each is paired with its archival reference. Switching versions restarts the excerpt; durations differ. Historical reference and voice research, not endorsement.
Anime
Original Japanese and auto-dubbed English. The system separates dialogue from the finished soundtrack rather than relying on clean dialogue stems, then translates and dubs the scene with its meaning, expression and background mix in focus.
Hardest stress test
Expressive Chinese-to-English singing combines a tonal source language with different target-language phrasing. The original Chinese enters at 0:28 and English at 0:58; the whole video alternates between them.
04 / Jeju TTS
Four Jeju speech samples from a model trained with just a few minutes of task-specific audio from one speaker. These examples are separate from the expressive English TTS performance.
05 / Expressive TTS
Use quality and expression tags or natural-language direction to guide delivery, including laughter, crying, whispers and shouts. This section contains the English performance. It is a recording, not a live prompt interface.
INTELLIGENT DYNAMICS

Contact for demo.

Oskar Jonientz · Co-founder

oskarjonientz@gmail.com
audiodynamics.dev
RELATED WORK

Preview

Want to see more?See more in slideshow ↗