← Back to projects

Internal research prototype · Case study

AudioGen

AudioGen connects voices, sound effects, music, and arrangement in a local workflow. I use it to explore how AI-generated audio can become a deliberately crafted scene while selection and control remain with the person creating it.

My role
Concept, desktop development, and AI integration
Technical foundation
Python · PySide6 · Local AI
Status
Internal prototype, tested in practice

01 / Production test

One workflow. An audible result.

NOIR CALL is a short audio drama with two voices, a telephone effect, rain, city sounds, and music. It served as a complete production test: generate material, select takes, process them, and arrange multiple tracks into a scene.

NOIR CALL

Final test mix · 66 seconds · English dialogue

AudioGen Composer with separate tracks for voices, effects, and music
The Composer in the documented prototype: voices, ambience, and effects are arranged on separate tracks for NOIR CALL. Select the image to enlarge it.

02 / Workflow

From individual takes to a crafted scene.

The core development challenge sits between the generators: variants need to remain comparable, selected takes need processing, and the results need to stay easy to find in the arrangement.

  1. 01

    Generate and select

    Dialogue lines are assigned to speakers and generated as individual takes. Variants can be reviewed and replaced without rebuilding the entire scene.

  2. 02

    Review and process

    Effects shape the sound and space. Experimental audio analysis helps identify unusual recordings. It does not replace listening or human approval.

  3. 03

    Bring it together

    Voices, effects, and music meet in a shared timeline. Timing, levels, and transitions turn individual files into a coherent production.

More views of the documented prototype
AudioGen Script view with dialogue lines and speaker assignments
Dialogue work at the level of individual lines and takes.
AudioGen FX Studio with an effect chain for a voice
Voice processing, shown here with an effect chain for the telephone sound.

03 / Insights

What the production test showed.

The connections matter

NOIR CALL demonstrates that generation, selection, processing, and arrangement can work together in a local workflow. For my work, that connection is more valuable than the number of integrated models.

A prototype is still a prototype

One completed scene is not a general quality benchmark. Voices and models behave differently, and technical scores remain signals for review. This case study documents the tested research work, not a claim of product readiness.

← Back to projects