Crate blazen_audio_stt 

blazen-audio-stt

Multi-backend speech-to-text engine crate for Blazen. Sibling to blazen-audio-tts, blazen-audio-music, and blazen-audio-codec; all four share the capability-agnostic vocabulary defined in blazen_audio.

Surface shape

• SttBackend: capability trait extending blazen_audio::AudioBackend that every STT engine implements.
• SttBackendHandle<B>: typed wrapper for Rust callers; monomorphizes on the concrete backend.
• DynSttProvider: erased wrapper (Box<dyn SttBackend>) for FFI / language-binding boundaries that cannot carry generics. See Appendix B of the PR-AUDIO plan for the dual-shape rationale.
• SttOptions: cross-backend options (model id, language hint, sample rate, device, diarization toggle).
• SttError: capability-agnostic error type; flattens engine-native failures into one of a small set of variants and implements From<SttError> for blazen_audio::AudioError.
• TranscriptionResult / TranscriptionSegment: result types.
• StreamingTranscript: per-chunk emission from the streaming SttBackend::stream surface.

Backends

Each backend lives in backends under its own feature gate:

Backend	Feature	Notes
whisper.cpp	whispercpp	Local CPU/GPU via the whisper-rs bindings.
candle	candle	Pure-Rust Whisper via candle-transformers.

Platform-specific acceleration for the whisper.cpp backend (cuda, metal, coreml) is exposed as opt-in no-op alias features — consumers wanting GPU acceleration must add whisper-rs as a direct dependency in their binary crate. See this crate’s Cargo.toml comments for the rationale.

Re-exports

pub use error::SttError;

pub use options::SttOptions;

pub use provider::DynSttProvider;

pub use provider::SttBackendHandle;

pub use traits::StreamingTranscript;

pub use traits::SttBackend;

pub use traits::TranscriptionResult;

pub use traits::TranscriptionSegment;

Modules

backends

Concrete SttBackend implementations.

error

Error type for the SttBackend trait surface and backend implementations.

options

Shared options for SttBackend implementations.

provider

Dual-shape provider wrappers around an SttBackend.

traits

The SttBackend trait — capability extension of blazen_audio::AudioBackend for speech-to-text engines.