Extending plantain2asr

plantain2asr is designed to be extended. Every component - models, normalizers, metrics, report sections - is a subclass of a simple abstract base class.

If you are only trying to run comparisons or export research artifacts, stop here and use the Interactive Constructor or Experiment first. This section is for custom component authors.

graph TD
    Processor["Processor (ABC)"]
    Processor --> BaseNormalizer
    Processor --> BaseASRModel
    Processor --> BaseMetric
    BaseSection["BaseSection (ABC)"]

    BaseNormalizer --> SimpleNormalizer
    BaseNormalizer --> DagrusNormalizer
    BaseNormalizer --> YourNormalizer["✨ YourNormalizer"]

    BaseASRModel --> GigaAM
    BaseASRModel --> Whisper
    BaseASRModel --> YourModel["✨ YourModel"]

    BaseMetric --> WER
    BaseMetric --> CER
    BaseMetric --> YourMetric["✨ YourMetric"]

    BaseSection --> MetricsSection
    BaseSection --> ErrorFrequencySection
    BaseSection --> YourSection["✨ YourSection"]

---

The rule: implement the interface, get the pipeline

Every component that subclasses Processor automatically works with >>:

dataset >> YourNormalizer()   # ✅ works
dataset >> YourModel()        # ✅ works
dataset >> YourMetric()       # ✅ works

---

Four extension points

What to add	Base class	Guide
Text normalization rules	BaseNormalizer	Custom Normalizer
A new ASR model	BaseASRModel	Custom Model
A new quality metric	BaseMetric	Custom Metric
A new report tab	BaseSection	Custom Report Section

Tip

Start with the guide for the component type you need. Each guide has a minimal example you can copy and adapt.