src/time_machine/application/speech_orchestrator.py

from __future__ import annotations

from collections.abc import Iterable, Iterator

from time_machine.application.encounter_service import EncounterService
from time_machine.domain.events import ErrorEvent, TemporalEvent
from time_machine.ports.speech import STTAdapter


class SpeechOrchestrator:
    def __init__(self, stt: STTAdapter, encounter_service: EncounterService) -> None:
        self.stt = stt
        self.encounter_service = encounter_service

    def handle_audio_turn(
        self,
        encounter_id: str,
        audio: object,
    ) -> Iterator[TemporalEvent]:
        transcript = self.stt.transcribe(audio)
        if not transcript.text.strip():
            yield ErrorEvent(
                message="No speech could be decoded from the signal.",
                severity="recoverable",
                metadata={"encounter_id": encounter_id},
            )
            return
        yield from self.encounter_service.handle_text_turn(
            encounter_id=encounter_id,
            text=transcript.text.strip(),
        )

    def handle_audio_stream(
        self,
        encounter_id: str,
        audio_chunks: Iterable[bytes],
    ) -> Iterator[TemporalEvent]:
        final_text = ""
        for event in self.stt.stream_transcript(audio_chunks):
            yield event
            if event.transcript.is_final:
                final_text = event.transcript.text

        final_text = final_text.strip()
        if final_text:
            yield from self.encounter_service.handle_text_turn(
                encounter_id=encounter_id,
                text=final_text,
            )
        else:
            yield ErrorEvent(
                message="No complete speech turn arrived from the signal.",
                severity="recoverable",
                metadata={"encounter_id": encounter_id},
            )