Models

Models factory

from plantain2asr import Models

The factory is the main entry point for built-in backends.

Supported calls:

Call	Backend	Extra	Device
Models.GigaAM_v3()	GigaAM v3 e2e-RNNT	gigaam	CUDA / MPS / CPU
Models.GigaAM_v3(model_name="e2e_ctc")	GigaAM v3 e2e-CTC	gigaam	CUDA / MPS / CPU
Models.GigaAM_v3(model_name="rnnt")	GigaAM v3 RNNT	gigaam	CUDA / MPS / CPU
Models.GigaAM_v3(model_name="ctc")	GigaAM v3 CTC	gigaam	CUDA / MPS / CPU
Models.GigaAM_v2(model_name="v2_rnnt")	GigaAM v2 RNNT	gigaam	CUDA / MPS / CPU
Models.GigaAM_v2(model_name="v2_ctc")	GigaAM v2 CTC	gigaam	CUDA / MPS / CPU
Models.Whisper()	Whisper large-v3 RU	whisper	CUDA / MPS / CPU
Models.Tone()	T-one	tone + T-One source archive	CUDA / CPU
Models.Vosk(model_path=...)	Vosk	vosk	CPU
Models.Canary()	NVIDIA Canary	canary	CUDA
Models.SaluteSpeech()	SaluteSpeech API	none	cloud

The factory also supports flexible name resolution:

model = Models.create("gigaam-v3")
model = Models.create("GigaAM_v3")
model = Models.create("tone")

Unknown names raise a helpful error with close suggestions.

For T-One, install the runtime extra first and then the source archive:

pip install plantain2asr[tone]
pip install "tone @ https://github.com/voicekit-team/T-one/archive/3c5b6c015038173840e62cea99e10cdb1c759116.tar.gz"

Usage

ds >> Models.GigaAM_v3()
ds >> Models.Whisper()

Model outputs are cached automatically and can be reused across later metric, report, and export steps.

BaseASRModel

from plantain2asr.models.base import BaseASRModel

class BaseASRModel(ABC):
    @property
    def name(self) -> str: ...

    def transcribe(self, audio_path: str) -> str: ...
    def transcribe_batch(self, paths: list) -> list: ...

    @property
    def is_e2e(self) -> bool: ...

Training-capable models can additionally expose training metadata used by the training layer.

See Custom Model for extension patterns.