import type { SpeechToText, SpeechToTextStartOptions } from "./types.js";

interface SpeechRecognitionResultEvent extends Event {
  readonly results: ArrayLike<{
    readonly 0: { readonly transcript: string };
  }>;
}

interface SpeechRecognitionInstance {
  continuous: boolean;
  interimResults: boolean;
  lang: string;
  onend: (() => void) | null;
  onerror: (() => void) | null;
  onresult: ((event: SpeechRecognitionResultEvent) => void) | null;
  start(): void;
  stop(): void;
}

interface SpeechRecognitionConstructor {
  new (): SpeechRecognitionInstance;
}

type SpeechRecognitionWindow = Window & {
  SpeechRecognition?: SpeechRecognitionConstructor;
  webkitSpeechRecognition?: SpeechRecognitionConstructor;
};

/**
 * Adapts standard and WebKit-prefixed recognition into the small capability
 * consumed by SurfaceView, without exposing browser-specific event shapes.
 */
export function createSpeechToText(): SpeechToText {
  const speechWindow = window as SpeechRecognitionWindow;
  const Recognition =
    speechWindow.SpeechRecognition ?? speechWindow.webkitSpeechRecognition;

  return {
    isSupported: Recognition !== undefined,
    start(options: SpeechToTextStartOptions) {
      if (Recognition === undefined) return undefined;

      const recognition = new Recognition();
      recognition.continuous = false;
      recognition.interimResults = false;
      recognition.lang = document.documentElement.lang || "it-IT";
      // Only the final first alternative is used; interim transcripts would
      // repeatedly overwrite text the visitor may already be editing.
      recognition.onresult = (event) => {
        const transcript = event.results[0]?.[0]?.transcript.trim();
        if (transcript) options.onTranscript(transcript);
      };
      recognition.onerror = () => options.onError();
      recognition.onend = () => options.onEnd();
      recognition.start();
      return { stop: () => recognition.stop() };
    },
  };
}