Expand description
Who is speaking when: the contract every speech-segmentation engine implements.
An embedder answers “whose voice is this window” and needs the window to hold one person. A segmenter answers the question underneath: given a fixed stretch of audio, which voices are active in each frame of it, and where does one stop and the next begin. Two people talking at once is an ordinary answer here rather than a window nobody can embed.
What a caller does with those frames — masking audio per voice, grouping the voices across a recording, naming them — is policy, not an engine’s business.
Structs§
- Activity
- Which of a chunk’s local voices speak in one frame.
- Chunk
Shape - The grid one engine answers on.
Enums§
Traits§
- Engine
- How an engine is constructed from its own configuration. Implementing
this beside
Segmenteris the whole checklist for a new engine. - Segmenter
- Speakers active in each frame of one fixed-width chunk of audio.
Functions§
- validate_
model_ file Non- target_family=wasm - The shared model validator, answering in this capability’s vocabulary.