Skip to content

Evaluate local UniDic (fugashi/Sudachi) for in-process accent + lower latency #50

Description

@torrid-fish

Problem

MarkAccent tokenises via a Yahoo MA HTTP round-trip, which adds per-chunk network latency and a runtime dependency on an external service. Evaluate replacing it with a local, in-process tokeniser (fugashi / SudachiPy plus a UniDic dictionary) to cut latency and drop the network dependency, and check whether UniDic's per-morpheme accent kernel (アクセント型 / aType) is usable as an accent signal.

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions