Module agents.stt

class FallbackSTT(STT, FallbackBase):
    def __init__(self, providers: List[STT], temporary_disable_sec: float = 60.0, permanent_disable_after_attempts: int = 3):
        STT.__init__(self)
        FallbackBase.__init__(self, providers, "STT", temporary_disable_sec=temporary_disable_sec, permanent_disable_after_attempts=permanent_disable_after_attempts)
        
        self._transcript_callback = None
        self._setup_event_listeners()

    def _setup_event_listeners(self):
        """Attach error listener to the currently active provider."""
        self.active_provider.on("error", self._on_provider_error)

    def _on_provider_error(self, error_msg):
        """Handle async errors (e.g. WebSocket disconnects)"""
        failed_p = self.active_provider
        asyncio.create_task(self._handle_async_error(str(error_msg), failed_p))

    async def _handle_async_error(self, error_msg: str, failed_provider: Any):
        """Async wrapper to handle switching logic"""
        switched = await self._switch_provider(f"Async Error: {error_msg}", failed_provider=failed_provider)
        if not switched:
            self.emit("error", error_msg)

    async def _switch_provider(self, reason: str, failed_provider: Any = None):
        """Override switch to handle STT specific setup"""
        provider_to_cleanup = failed_provider if failed_provider else self.active_provider

        try:
            provider_to_cleanup.off("error", self._on_provider_error)
        except: pass

        active_before = self.active_provider
        switched = await super()._switch_provider(reason, failed_provider)
        active_after = self.active_provider
        
        if switched:
            if active_before != active_after:
                if self._transcript_callback:
                    self.active_provider.on_stt_transcript(self._transcript_callback)
                self.active_provider.on("error", self._on_provider_error)
            return True
        return False

    def on_stt_transcript(self, callback) -> None:
        """Capture the callback so we can re-apply it after switching."""
        self._transcript_callback = callback
        self.active_provider.on_stt_transcript(callback)

    async def process_audio(self, audio_frames: bytes, **kwargs) -> None:
        """
        Main entry point. If this fails, it's usually a connection error.
        We catch, switch, and retry immediately.
        """
        if self.check_recovery():
            if self._transcript_callback:
                self.active_provider.on_stt_transcript(self._transcript_callback)
            self.active_provider.on("error", self._on_provider_error)

        current_provider = self.active_provider
        try:
            await current_provider.process_audio(audio_frames, **kwargs)
        except Exception as e:
            switched = await self._switch_provider(str(e), failed_provider=current_provider)
            if switched:
                await self.active_provider.process_audio(audio_frames, **kwargs)
            else:
                raise e

    async def aclose(self) -> None:
        """Close all providers."""
        for p in self.providers:
            await p.aclose()
        await super().aclose()

class STT(EventEmitter[Literal["error"]]):
    """Base class for Speech-to-Text implementations"""
    
    def __init__(
        self,
    ) -> None:
        super().__init__()
        self._label = f"{type(self).__module__}.{type(self).__name__}"
        self._transcript_callback: Optional[Callable[[STTResponse], Awaitable[None]]] = None
        
    @property
    def label(self) -> str:
        """Get the STT provider label"""
        return self._label

    def on_stt_transcript(self, callback: Callable[[STTResponse], Awaitable[None]]) -> None:
        """Set callback for receiving STT transcripts"""
        self._transcript_callback = callback

    @abstractmethod
    async def process_audio(
        self,
        audio_frames: bytes,
        language: Optional[str] = None,
        **kwargs: Any
    ) -> None:
        """
        Process audio frames and convert to text
        
        Args:
            audio_frames: Iterator of bytes to process
            language: Optional language code for recognition
            **kwargs: Additional provider-specific arguments
            
        Returns:
            AsyncIterator yielding STTResponse objects
        """
        raise NotImplementedError

    async def aclose(self) -> None:
        """Cleanup resources"""
        logger.info(f"Cleaning up STT: {self.label}")
        self._transcript_callback = None
        try:
            import gc
            gc.collect()
            logger.info(f"STT garbage collection completed: {self.label}")
        except Exception as e:
            logger.error(f"Error during STT garbage collection: {e}")
        
        logger.info(f"STT cleanup completed: {self.label}")
    
    async def __aenter__(self) -> STT:
        return self
        
    async def __aexit__(self, exc_type, exc_val, exc_tb) -> None:
        await self.aclose()

class STTResponse(BaseModel):
    """Response from STT processing
    
    Attributes:
        event_type: The type of speech event.
        data: The data from the speech event.
        metadata: Additional metadata from the speech event.
    """
    event_type: SpeechEventType
    data: SpeechData
    metadata: Optional[dict[str, Any]] = None

@dataclass
class SpeechData:
    """Data structure for speech recognition results
    
    Attributes:
        text: The recognized text.
        confidence: The confidence level of the recognition.
        language: The language of the recognized text.
        start_time: The start time of the speech.
        end_time: The end time of the speech.
    """
    text: str
    confidence: float = 0.0
    language: Optional[str] = None
    start_time: float = 0.0
    end_time: float = 0.0

class SpeechEventType(str, Enum):
    """Type of speech event"""
    START = "start_of_speech"
    INTERIM = "interim_transcript"
    PREFLIGHT = "preflight_transcript"
    FINAL = "final_transcript"
    END = "end_of_speech"