Crate atomr_agents_stt_core 

Core types for the atomr-agents speech-to-text capability.

This crate is intentionally I/O-free: it defines the SpeechToText and StreamingSession traits, the rich Capabilities struct that backends advertise via a pub const, the audio-input and transcript data types, and a deterministic MockSpeechToText for tests.

Concrete backends live in sibling crates:

• atomr-agents-stt-runtime-openai — OpenAI Whisper REST.
• atomr-agents-stt-runtime-deepgram — Deepgram REST + WS.
• atomr-agents-stt-runtime-assemblyai — AssemblyAI REST + WS.
• atomr-agents-stt-runtime-whisper — local whisper-rs.

Audio I/O (symphonia, cpal) lives in atomr-agents-stt-audio, the higher-level voice-session abstraction in atomr-agents-stt-voice, and the agent-framework adapters in atomr-agents-stt-tool.

Structs

Capabilities

Serialize-only by design: the slice fields (languages and supported_audio_formats) are &'static, which serde can’t deserialize into. Capabilities flow outward (to JSON / Python / telemetry) and are never round-tripped back into Rust.

MockSpeechToText

Deterministic mock STT. transcribe returns a transcript whose text is a hash digest of the input length so tests can assert on stability without caring about the exact string.

PcmBuffer

Decoded PCM. Backends that resample (e.g. whisper-rs needs 16 kHz mono f32) take this and convert.

Segment

SpeakerTag

StreamOptions

Per-call options for opening a streaming session. Mirrors the “common knobs” across the four MVP backends.

TranscribeOptions

Per-call options shared across batch transcription backends.

Transcript

Word

Enums

AudioFormat

AudioInput

BackendKind

DiarizationSupport

Languages

SampleType

StreamEvent

SttError

TransportKind

Traits

SpeechToText

Speech-to-text backend. Implementations live in sibling stt-runtime-* crates.

StreamingSession

Active streaming session. Caller alternates push_audio/finish with consuming the stream returned from events.

Type Aliases

DynSpeechToText

Result