Practical overview
When it fits
Use VOCA when you have a speech recording and a static character mesh and need a generated facial animation sequence, especially when mesh output will be edited or rendered in a further workflow.
Who it is for
- 3D character creators who want to drive facial animation from recorded speech.
- Developers and researchers exploring speech-conditioned facial motion synthesis.
- Users who want mesh sequences that can be edited with FLAME or rendered to video.
Documented uses
- Generate animated character meshes from an audio speech signal and static template mesh.
- Render a generated animation sequence to video.
- Demonstrate edits to FLAME-based identity shape and head pose.
- Generate character templates using the FLAME identity shape space or a template reconstructed from an image using RingNet.
Requirements & limits
- The documented demo depends on separate trained models and assets, including VOCA, FLAME, and DeepSpeech 0.1.0.
- The README identifies Python 3.6.8 and TensorFlow 1.14.0, but its setup commands use Python 3.7 for the virtual environment.
- Default pyrender visualization may be problematic in some configurations; the README documents saving meshes with visualization disabled.
- The eye-blink example is limited: it demonstrates eyelid manipulation and does not produce motion resembling a true eye blink.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Set up the VOCA demo environment
Prepare the documented Python environment and fetch the assets required by the demo.
Before you start
- Linux commands and Python environment tools described by the README
- pip and virtualenv
- ffmpeg
- Mesh-processing libraries from MPI-IS/mesh
- Trained VOCA model, audio sequences, template meshes, FLAME model, and DeepSpeech 0.1.0 model
Steps
Clone the VOCA repository and create and activate a virtual environment.
Install the README-specified pip version, ffmpeg, MPI-IS/mesh libraries, and the project requirements.
Run the documented data-preparation script after obtaining the required model and example assets.
Expected outcome
The documented environment and demo data are prepared for the next stage of running a VOCA demo.
Guide limitations
- The README specifies Python 3.6.8 as used and TensorFlow 1.14.0 as tested, while its setup commands create a Python 3.7 virtual environment.
- This is setup guidance only; it does not verify installation or execution.
Generate and render a speech-driven animation
Use the documented demo command to produce animated meshes from an audio file and template mesh, then render the sequence.
Before you start
- VOCA environment and required demo assets
- A speech audio file and a static template mesh
- Trained VOCA model and DeepSpeech model files
Steps
Run the README’s
run_voca.pydemo with the model, DeepSpeech, audio, template, condition index, and output-path arguments.Use the documented
visualize_sequence.pycommand with the generated mesh sequence and audio file to render a video.If default visualization causes problems in the environment, use the documented
--visualize Falseoption to save meshes without that visualization.
Expected outcome
The demo produces an animated mesh sequence and can render it to video; disabling visualization leaves the meshes available for another viewing or rendering tool.
Guide limitations
- The README notes that pyrender visualization can have problems in certain configurations, including remote use.
- The README’s commands are documentation, not execution verification.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 18 October 2026
- Original website title
- TimoBolkart/voca · Official homepage title · retained in its source language
Related options
VOCA Blender Addon — A separately linked Blender add-on related to VOCA; the repository does not describe it as part of this framework.