@wirecat/bitrix24-cli 0.0.0-stage → 0.1.0

package/dist/bundle/chunk-6WY72XVQ.js

@@ -0,0 +1,216 @@
+import { createRequire as __bundleRequire } from "node:module";
+const require = __bundleRequire(import.meta.url);
+import {
+  CliError
+} from "./chunk-IRV2PCQ2.js";
+import {
+  importFrom
+} from "./chunk-Y5E2WNWB.js";
+import {
+  __name
+} from "./chunk-ZVVMGPBJ.js";
+
+// node_modules/.pnpm/@wirecat+cli-core@0.586.0_@wirecat+cli-base@0.79.0_commander@15.0.0_esbuild@0.28.2_type_bdecddef0ae5a50ae90cab8b9e35c13b/node_modules/@wirecat/cli-core/dist/speech/recognize.js
+import { createRequire } from "node:module";
+
+// node_modules/.pnpm/@wirecat+cli-core@0.586.0_@wirecat+cli-base@0.79.0_commander@15.0.0_esbuild@0.28.2_type_bdecddef0ae5a50ae90cab8b9e35c13b/node_modules/@wirecat/cli-core/dist/speech/ogg-limit.js
+var invalid = /* @__PURE__ */ __name(() => new CliError("validation_error", "local transcription needs a complete mono or stereo Ogg Opus recording"), "invalid");
+var checkOggDuration = /* @__PURE__ */ __name((bytes) => {
+  const view = new DataView(bytes.buffer, bytes.byteOffset, bytes.byteLength);
+  let offset = 0;
+  let serial;
+  let packet = 0;
+  let length = 0;
+  const prefix = [];
+  while (offset < bytes.length) {
+    if (offset + 27 > bytes.length || view.getUint32(offset, false) !== 1332176723 || bytes[offset + 4] !== 0)
+      throw invalid();
+    const stream = view.getUint32(offset + 14, true);
+    if (serial !== void 0 && serial !== stream)
+      throw invalid();
+    serial = stream;
+    const segments = bytes[offset + 26] ?? 0;
+    let data = offset + 27 + segments;
+    if (data > bytes.length || Boolean((bytes[offset + 5] ?? 0) & 1) !== length > 0)
+      throw invalid();
+    for (let index = 0; index < segments; index++) {
+      const size = bytes[offset + 27 + index] ?? 0;
+      if (data + size > bytes.length)
+        throw invalid();
+      const take = Math.min(size, 19 - prefix.length);
+      for (let i = 0; i < take; i++)
+        prefix.push(bytes[data + i] ?? 0);
+      length += size;
+      data += size;
+      if (size === 255)
+        continue;
+      if (packet === 0) {
+        if (length < 19 || String.fromCharCode(...prefix.slice(0, 8)) !== "OpusHead" || ![1, 2].includes(prefix[9] ?? 0) || prefix[18] !== 0)
+          throw invalid();
+      } else if (packet === 1) {
+        if (length < 8 || String.fromCharCode(...prefix.slice(0, 8)) !== "OpusTags")
+          throw invalid();
+      } else {
+        if (length < 1)
+          throw invalid();
+        const toc = prefix[0] ?? 0;
+        const config = toc >> 3;
+        const frame = config >= 16 ? 120 << (config & 3) : config >= 12 ? 480 << (config & 1) : (config & 3) === 3 ? 2880 : 480 << (config & 3);
+        const count = (toc & 3) === 0 ? 1 : (toc & 3) === 3 ? length >= 2 ? (prefix[1] ?? 0) & 63 : 0 : 2;
+        if (count < 1 || frame * count > 5760)
+          throw invalid();
+      }
+      packet++;
+      length = 0;
+      prefix.length = 0;
+    }
+    offset = data;
+  }
+  if (length !== 0 || packet < 3)
+    throw invalid();
+}, "checkOggDuration");
+
+// node_modules/.pnpm/@wirecat+cli-core@0.586.0_@wirecat+cli-base@0.79.0_commander@15.0.0_esbuild@0.28.2_type_bdecddef0ae5a50ae90cab8b9e35c13b/node_modules/@wirecat/cli-core/dist/speech/recognize.js
+var SAMPLE_RATE = 16e3;
+var LEAD = SAMPLE_RATE / 2;
+var toModelRate = /* @__PURE__ */ __name((samples, rate) => {
+  const step = rate / SAMPLE_RATE;
+  const length = Math.floor(samples.length / step);
+  const pcm = new Float32Array(LEAD + length);
+  for (let i = 0; i < length; i++) {
+    const from = Math.floor(i * step);
+    const to = Math.max(from + 1, Math.floor((i + 1) * step));
+    let sum = 0;
+    for (let j = from; j < to; j++)
+      sum += samples[j] ?? 0;
+    pcm[LEAD + i] = sum / (to - from);
+  }
+  return pcm;
+}, "toModelRate");
+var decodeOgg = /* @__PURE__ */ __name(async (bytes) => {
+  checkOggDuration(bytes);
+  const { OggOpusDecoder } = await importFrom(new URL("speech/recognize.js", import.meta.resolve("@wirecat/cli-core")).href, "ogg-opus-decoder");
+  const decoder = new OggOpusDecoder();
+  await decoder.ready;
+  try {
+    const { channelData, sampleRate, samplesDecoded } = await decoder.decodeFile(bytes);
+    if (!channelData[0] || samplesDecoded === 0)
+      throw new Error("no Ogg Opus audio in it");
+    return channelData.map((samples) => ({ samples, rate: sampleRate }));
+  } finally {
+    decoder.free();
+  }
+}, "decodeOgg");
+var WINDOW = 512;
+var detector = /* @__PURE__ */ __name((vadModel) => ({
+  sileroVad: {
+    model: vadModel,
+    // At the default 0.5 a quietly spoken stretch in the middle of a voice message never counted as
+    // speech and was lost; 0.3 keeps it for both models and changes nothing on clear speech (126508, 2026-09-30).
+    threshold: 0.3,
+    minSilenceDuration: 0.5,
+    minSpeechDuration: 0.25,
+    maxSpeechDuration: 20,
+    windowSize: WINDOW
+  },
+  sampleRate: SAMPLE_RATE,
+  numThreads: 1,
+  provider: "cpu",
+  debug: 0,
+  bufferSizeInSeconds: 30
+}), "detector");
+var cut = /* @__PURE__ */ __name((vad, pcm, each) => {
+  vad.reset();
+  const last = Math.max(0, pcm.length - LEAD);
+  const drain = /* @__PURE__ */ __name(() => {
+    for (; !vad.isEmpty(); vad.pop()) {
+      const { start, samples } = vad.front();
+      each({
+        start: Math.min(last, Math.max(0, start - LEAD)) / SAMPLE_RATE,
+        end: Math.min(last, Math.max(0, start + samples.length - LEAD)) / SAMPLE_RATE,
+        samples
+      });
+    }
+  }, "drain");
+  for (let i = 0; i + WINDOW <= pcm.length; i += WINDOW) {
+    vad.acceptWaveform(pcm.subarray(i, i + WINDOW));
+    drain();
+  }
+  vad.flush();
+  drain();
+}, "cut");
+var openSpeechCutter = /* @__PURE__ */ __name((vadModel) => {
+  const vad = loadSherpa().createVad(detector(vadModel));
+  return {
+    cut: /* @__PURE__ */ __name((pcm) => {
+      const spans = [];
+      cut(vad, pcm, ({ start, end }) => spans.push({ start, end }));
+      return spans;
+    }, "cut"),
+    free: /* @__PURE__ */ __name(() => vad.free(), "free")
+  };
+}, "openSpeechCutter");
+var loaded;
+var loadSherpa = /* @__PURE__ */ __name(() => {
+  if (loaded)
+    return loaded;
+  const load = createRequire(new URL("speech/recognize.js", import.meta.resolve("@wirecat/cli-core")).href);
+  const wasm = {};
+  load("sherpa-onnx/sherpa-onnx-wasm-nodejs.js")(wasm);
+  const stdout = wasm.FS?.getStream(1);
+  if (stdout)
+    stdout.nfd = 2;
+  const asr = load("sherpa-onnx/sherpa-onnx-asr.js");
+  const vad = load("sherpa-onnx/sherpa-onnx-vad.js");
+  loaded = {
+    createOfflineRecognizer: /* @__PURE__ */ __name((config) => new asr.OfflineRecognizer(config, wasm), "createOfflineRecognizer"),
+    createVad: /* @__PURE__ */ __name((config) => vad.createVad(wasm, config), "createVad")
+  };
+  return loaded;
+}, "loadSherpa");
+var openRecognizer = /* @__PURE__ */ __name((model, path, vadModel) => {
+  const sherpa = loadSherpa();
+  const engine = sherpa.createOfflineRecognizer({
+    featConfig: { sampleRate: SAMPLE_RATE, featureDim: model.featureDim },
+    modelConfig: { ...model.config(path), numThreads: 1, provider: "cpu", debug: 0 },
+    decodingMethod: "greedy_search"
+  });
+  const vad = sherpa.createVad(detector(vadModel));
+  const piece = /* @__PURE__ */ __name((samples) => {
+    const stream = engine.createStream();
+    try {
+      stream.acceptWaveform(SAMPLE_RATE, samples);
+      engine.decode(stream);
+      return engine.getResult(stream).text.trim();
+    } finally {
+      stream.free();
+    }
+  }, "piece");
+  const pieces = /* @__PURE__ */ __name((pcm) => {
+    const heard = [];
+    cut(vad, pcm, ({ start, end, samples }) => {
+      const text = piece(samples);
+      if (text)
+        heard.push({ start, end, text });
+    });
+    return heard;
+  }, "pieces");
+  return {
+    recognize: /* @__PURE__ */ __name((pcm) => pieces(pcm).map(({ text }) => text).join(" "), "recognize"),
+    pieces,
+    free: /* @__PURE__ */ __name(() => {
+      vad.free();
+      engine.free();
+    }, "free")
+  };
+}, "openRecognizer");
+
+export {
+  SAMPLE_RATE,
+  toModelRate,
+  decodeOgg,
+  detector,
+  openSpeechCutter,
+  loadSherpa,
+  openRecognizer
+};