FunASR/docs/reference/build_task.md

Build custom tasks

This is the migration entry for the former Task-based tutorial. That tutorial described an older FunASR architecture; its AbsTask / ASRTask examples are not interfaces in the current checkout. Do not copy them into a new integration.

Choose an extension point

What you need	Maintained guide	What to verify
Run an existing checkpoint	Python SDK	Model, input and output contract
Fine-tune an existing architecture	Training	Recipe, data format, checkpoint and evaluation
Add a model architecture	Model registration	Registry name, configuration, inference() and result fields
Connect a third-party model	MOSS integration	Upstream attribution, dependencies and adapter boundary
Expose a transcription service	Deployment matrix	Runtime, protocol and supported model combination

The current model registration mechanism is implemented in register.py; inference orchestration is in auto_model.py. Registering a model is not the same as adding support to every HTTP, WebSocket or C++ runtime. Follow the specific service contract and test the intended route separately.

Migrate an older extension

1. Record the original FunASR commit, configuration, checkpoint and dependency versions before changing the environment.
2. Choose a current recipe or registered model with matching inputs and outputs. Map the old preprocessing, collation and model construction responsibilities to that recipe; do not assume the old Task class has a drop-in replacement.
3. Test configuration loading, checkpoint compatibility and a small inference sample before training or serving. Compare outputs against the old pinned environment, then run the relevant evaluation.

For historical experiments, keep their matching source revision in an isolated environment. This page preserves the old documentation URL; current implementation instructions live in the guides above.