STT

import { defineModels } from "@scrydon/sdk-authoring/extensions/define";
import type { STTCapabilityConfig } from "@scrydon/sdk-authoring/extensions/capabilities/stt";
import type { SttStreamingSpec } from "@scrydon/sdk-authoring/extensions/capabilities/stt-streaming";

const LIVE_STREAMING: SttStreamingSpec = {
  protocol: "acme-live-v1",
  audio: { sampleRateHz: 16000, encoding: "pcm16", channels: 1 },
};

const sttCapability: STTCapabilityConfig = {
  models: [
    { id: "whisper-v3", name: "Whisper V3", benchmarks: [{ name: "WER", score: 4.2, source: "Artificial Analysis" }] },
    { id: "live-v1", name: "Live Transcribe", streaming: LIVE_STREAMING },
  ],
  defaultModel: "whisper-v3",

  // Batch mode — upload audio, get text back
  async transcribe(request) {
    const response = await fetch("https://api.example.com/v1/transcribe", {
      method: "POST",
      headers: { Authorization: `Bearer ${request.apiKey}` },
      body: request.audioData,
    });
    const data = await response.json();
    return { transcript: data.text, language: data.language, confidence: data.confidence };
  },

  // Streaming (optional). A model that declares `streaming` needs BOTH hooks:
  // createStreamingSession mints the ticket, openStreamingSession runs the session.
  async createStreamingSession(request) {
    return {
      ...LIVE_STREAMING,
      transport: {
        mode: "proxied",
        upstreamUrl: "wss://api.example.com/v1/live",
        upstreamHeaders: { Authorization: `Bearer ${request.apiKey}` },
      },
      diarizes: false,
    };
  },
  async openStreamingSession(request, context) {
    const socket = new WebSocket(`wss://api.example.com/v1/live?model=${encodeURIComponent(request.model)}`);
    await new Promise((resolve, reject) => {
      socket.addEventListener("open", resolve, { once: true });
      socket.addEventListener("error", reject, { once: true });
    });
    // Authenticate the way your vendor's protocol expects.
    socket.send(JSON.stringify({ type: "auth", apiKey: request.apiKey }));
    context.signal.addEventListener("abort", () => socket.close());
    return {
      sendAudio: (pcm) => socket.send(new Uint8Array(pcm)),
      finalize: () => socket.send(JSON.stringify({ type: "finalize" })),
      close: () => socket.close(),
      onTranscript: (handler) =>
        socket.addEventListener("message", (event) => {
          const data = JSON.parse(String(event.data));
          handler({ text: data.text, isFinal: data.final === true });
        }),
      onError: (handler) => socket.addEventListener("error", () => handler(new Error("stream error"))),
      onClose: (handler) => socket.addEventListener("close", (event) => handler(event.reason || null)),
    };
  },
};

// One model family per runtime; list it under defineExtension()'s provides.models.
export const sttModels = defineModels({
  id: "my-service-stt",
  name: "My Service Speech-to-Text",
  family: "stt",
  runtime: sttCapability,
});