sophontalk-services 0.0.3 → 0.0.5

package/package.json

@@ -1,13 +1,12 @@
 {
   "name": "sophontalk-services",
-  "version": "0.0.3",
+  "version": "0.0.5",
   "type": "module",
   "scripts": {
     "dev": "vite",
     "build": "vue-tsc -b && vite build",
     "build:lib": "vite build --config vite.lib.config.ts",
-    "preview": "vite preview",
-    "publish": "npm publish --registry https://registry.npmjs.org --access public"
+    "preview": "vite preview"
   },
   "main": "./dist-lib/sophontalk-services.cjs",
   "module": "./dist-lib/sophontalk-services.js",

package/dist-lib/assets/feature.worker-Dx6moind.js

@@ -1,145 +0,0 @@
-import x from "fft.js";
-import { create as O, ConverterType as H } from "@alexanderolsen/libsamplerate-js";
-const y = 16e3, U = 25, D = 10, u = 80, S = 4, W = 0.97, L = 1e-10, b = 20, h = Math.round(U * y / 1e3), C = Math.round(D * y / 1e3), k = (e) => {
-  if (e <= 0) return 1;
-  let t = 1;
-  for (; t < e; ) t *= 2;
-  return t;
-}, P = k(h), N = new x(P), v = new Float32Array(h), G = () => {
-  if (h <= 1) return;
-  const e = 2 * Math.PI / (h - 1);
-  for (let t = 0; t < h; t++)
-    v[t] = 0.5 - 0.5 * Math.cos(t * e);
-};
-G();
-function Q(e, t, o, a = 20, s) {
-  s = s || o / 2;
-  const i = (n) => 700 * (Math.exp(n / 1127) - 1), r = (n) => 1127 * Math.log(1 + n / 700), l = r(a), A = r(s), w = new Float32Array(e + 2);
-  for (let n = 0; n < e + 2; n++)
-    w[n] = l + (A - l) * n / (e + 1);
-  const _ = w.map(i).map((n) => n * t / o), f = Math.floor(t / 2) + 1, m = [];
-  for (let n = 1; n <= e; n++) {
-    const c = _[n - 1], F = _[n], M = _[n + 1], I = Math.ceil(c), T = Math.floor(M), E = Math.max(0, I), p = Math.min(f - 1, T), B = p - E + 1;
-    if (B <= 0) {
-      m.push({ startBin: 0, values: new Float32Array(0) });
-      continue;
-    }
-    const R = new Float32Array(B);
-    for (let d = E; d <= p; d++) {
-      let g = 0;
-      d <= F ? g = (d - c) / (F - c) : g = (M - d) / (M - F), R[d - E] = g;
-    }
-    m.push({ startBin: E, values: R });
-  }
-  return m;
-}
-const q = Q(
-  u,
-  P,
-  y,
-  b
-);
-function K(e) {
-  for (let t = 0; t < e.length; t++)
-    e[t] *= v[t];
-}
-function Y(e, t) {
-  for (let o = e.length - 1; o > 0; o--)
-    e[o] -= t * e[o - 1];
-  e[0] -= t * e[0];
-}
-function V(e) {
-  const t = e.length;
-  if (t === 0) return;
-  let o = 0;
-  for (let s = 0; s < t; s++) o += e[s];
-  const a = o / t;
-  for (let s = 0; s < t; s++) e[s] -= a;
-}
-function Z(e, t) {
-  const o = e.length;
-  for (let a = 0; a < o; a++) {
-    const s = t[a * 2], i = t[a * 2 + 1];
-    e[a] = s * s + i * i;
-  }
-}
-function j(e) {
-  const t = e.length;
-  if (t < 2) return;
-  let o = 0;
-  for (let r = 0; r < t; r++) o += e[r];
-  const a = o / t;
-  let s = 0;
-  for (let r = 0; r < t; r++) {
-    const l = e[r] - a;
-    s += l * l;
-  }
-  const i = Math.sqrt(s / (t - 1));
-  if (i > 1e-8) {
-    const r = 1 / i;
-    for (let l = 0; l < t; l++)
-      e[l] = (e[l] - a) * r;
-  } else
-    for (let r = 0; r < t; r++)
-      e[r] -= a;
-}
-async function z(e) {
-  let t = e.leftChannel;
-  if (e.rightChannel) {
-    const f = e.rightChannel, m = new Float32Array(t.length);
-    for (let n = 0; n < t.length; n++)
-      m[n] = (t[n] + f[n]) * 0.5;
-    t = m;
-  }
-  if (e.sampleRate !== y) {
-    const f = await O(1, e.sampleRate, y, {
-      converterType: H.SRC_SINC_MEDIUM_QUALITY
-    });
-    t = f.simple(t), f.destroy();
-  }
-  const o = Math.floor((t.length - h) / C) + 1;
-  if (o <= 0)
-    throw new Error("特征提取失败：音频时长过短。");
-  const a = Math.ceil(o / S) * S, s = a * u, i = new Float32Array(s), r = new Float32Array(h), l = N.createComplexArray(), A = N.createComplexArray(), w = new Float32Array(Math.floor(P / 2) + 1);
-  for (let f = 0; f < o; f++) {
-    const m = f * C;
-    r.set(t.subarray(m, m + h)), V(r), Y(r, W), K(r), l.fill(0);
-    for (let c = 0; c < h; c++)
-      l[c * 2] = r[c];
-    N.transform(A, l), Z(w, A);
-    const n = f * u;
-    for (let c = 0; c < u; c++) {
-      const F = q[c];
-      let M = 0;
-      const I = F.values, T = F.startBin, E = I.length;
-      for (let p = 0; p < E; p++)
-        M += w[T + p] * I[p];
-      M = M < L ? L : M, i[n + c] = Math.log(M);
-    }
-  }
-  if (a > o) {
-    const f = (o - 1) * u, m = f + u, n = i.subarray(f, m);
-    for (let c = o; c < a; c++)
-      i.set(n, c * u);
-  }
-  j(i);
-  const _ = a / S;
-  return {
-    features: i,
-    dimensions: [_, S, u]
-  };
-}
-self.addEventListener(
-  "message",
-  async (e) => {
-    try {
-      const t = e.data, o = await z(t);
-      self.postMessage(
-        { status: "success", payload: o },
-        { transfer: [o.features.buffer] }
-      );
-    } catch (t) {
-      self.postMessage({ status: "error", error: t.message });
-    }
-  }
-);

package/dist-lib/assets/inference.worker-BPSCUMiL.js

@@ -1,178 +0,0 @@
-import { env as P, InferenceSession as $, Tensor as W } from "onnxruntime-web";
-import q from "jszip";
-const V = 80, G = 4, H = 25, j = 1e3 / H, J = (t) => {
-  if (t === 128) return 4;
-  if (t === 96) return 3;
-  throw new Error(`Unsupported crop size: ${t}. Only nano (96) and tiny (128) are supported.`);
-}, Q = (t, e, n) => {
-  if (e <= 1) return { nextIndex: 0, nextDirection: 1 };
-  let a = t + n, s = n;
-  return a >= e ? (a = e - 2, s = -1) : a < 0 && (a = 1, s = 1), { nextIndex: a, nextDirection: s };
-};
-P.wasm.numThreads = 1;
-const Z = () => typeof navigator < "u" && !!navigator.gpu, F = 32;
-let _ = null, R = null, x = null, L = null;
-const w = {
-  tensors: /* @__PURE__ */ new Map(),
-  faceImages: /* @__PURE__ */ new Map(),
-  fullImages: /* @__PURE__ */ new Map()
-};
-let S = 3, b = 96, z = 0, k = 0, C = null, A = null, y = null, E = null, g = null, I = null, D = null, M = null, T = !1, v = "", p = null;
-const B = (t) => {
-  self.postMessage({ type: "error", payload: t });
-}, K = (t) => {
-  const e = w.tensors.get(t);
-  if (!e) throw new Error(`Cache miss: ${t}`);
-  const n = b - 2 * S;
-  return new W("float32", e, [1, 6, n, n]);
-}, X = (t) => {
-  const e = w.faceImages.get(t);
-  if (!e) throw new Error(`Cache miss: ${t}`);
-  return e;
-}, Y = async (t) => {
-  const e = w.fullImages.get(t);
-  if (!e) throw new Error(`Cache miss: ${t}`);
-  return await createImageBitmap(e);
-}, ee = (t) => {
-  const [e, n, a, s] = t.dims, u = t.data;
-  (!y || y.width !== s || y.height !== a) && (C.width = s, C.height = a, y = A.createImageData(s, a));
-  const d = a * s, l = new Uint32Array(y.data.buffer), o = d, i = d * 2, c = n === 3;
-  for (let r = 0; r < d; r++) {
-    let m = (c ? u[r + i] : u[r]) * 255, f = (c ? u[r + o] : u[r]) * 255, h = u[r] * 255;
-    m = m < 0 ? 0 : m > 255 ? 255 : m, f = f < 0 ? 0 : f > 255 ? 255 : f, h = h < 0 ? 0 : h > 255 ? 255 : h, l[r] = 255 << 24 | (h & 255) << 16 | (f & 255) << 8 | m & 255;
-  }
-  A.putImageData(y, 0, 0);
-}, te = async (t, e, n) => {
-  ee(t);
-  const a = X(e.face_image);
-  g.globalCompositeOperation = "source-over", g.drawImage(a, 0, 0), g.drawImage(C, S, S), g.globalCompositeOperation = "destination-in", g.drawImage(L, 0, 0, b, b), g.globalCompositeOperation = "destination-over", g.drawImage(a, 0, 0), g.globalCompositeOperation = "source-over", D.drawImage(n, 0, 0);
-  const s = e.crop_info.ymax - e.crop_info.ymin;
-  return D.drawImage(
-    E,
-    e.crop_info.xmin,
-    e.crop_info.ymin,
-    e.crop_info.width,
-    s
-  ), await createImageBitmap(I);
-}, ae = (t, e, n) => {
-  const [a, s, u] = e, d = s * u, l = F * d;
-  (!M || M.length !== l) && (M = new Float32Array(l));
-  for (let o = 0; o < F; o++) {
-    const c = Math.min(
-      Math.max(n - F / 2 + o, 0),
-      a - 1
-    ) * d, r = o * d;
-    M.set(
-      t.subarray(c, c + d),
-      r
-    );
-  }
-  return M;
-}, ne = (t, e, n) => {
-  C || (C = new OffscreenCanvas(1, 1), A = C.getContext("2d", {
-    willReadFrequently: !0
-  })), (!E || E.width !== n) && (E = new OffscreenCanvas(n, n), g = E.getContext("2d", {
-    willReadFrequently: !0
-  })), (!I || I.width !== t || I.height !== e) && (I = new OffscreenCanvas(t, e), D = I.getContext("2d", {
-    willReadFrequently: !0
-  }));
-}, se = async (t) => {
-  try {
-    if (t.wasmPaths ? P.wasm.wasmPaths = t.wasmPaths : P.wasm.wasmPaths = "/", !t.dataset) throw new Error("Missing dataset");
-    const e = t.dataset;
-    R = e, b = e.dataset_info.config.crop_size, S = J(b), z = e.dataset_info.source_image_dimensions?.width || 0, k = e.dataset_info.source_image_dimensions?.height || 0, ne(z, k, b), x = await q.loadAsync(t.zipBuffer), t.zipBuffer = null, L = t.blendingMask;
-    const n = e.images.map(async (l) => {
-      if (!w.tensors.has(l.tensor_file)) {
-        const o = x.file(l.tensor_file);
-        if (o) {
-          const i = await o.async("arraybuffer");
-          w.tensors.set(l.tensor_file, new Float32Array(i));
-        }
-      }
-      if (!w.faceImages.has(l.face_image)) {
-        const o = x.file(l.face_image);
-        if (o) {
-          const i = await o.async("blob"), c = await createImageBitmap(i);
-          w.faceImages.set(l.face_image, c);
-        }
-      }
-      if (!w.fullImages.has(l.full_image)) {
-        const o = x.file(l.full_image);
-        if (o) {
-          const i = await o.async("blob");
-          w.fullImages.set(l.full_image, i);
-        }
-      }
-    });
-    await Promise.all(n), x = null;
-    const a = [];
-    Z() && a.push("webgpu"), a.push("wasm");
-    const s = new URL(t.modelPath, self.location.href).toString(), u = new URL(s).pathname.split("/").pop() || "model.onnx", d = new URL(`${u}.data`, s).toString();
-    _ = await $.create(s, {
-      executionProviders: a,
-      graphOptimizationLevel: "all",
-      enableMemPattern: !0,
-      enableCpuMemArena: !0,
-      externalData: [{ data: d, path: `${u}.data` }]
-    }), self.postMessage({ type: "ready" });
-  } catch (e) {
-    B(e?.message || "Init Error");
-  }
-}, oe = async (t) => {
-  if (!_ || !R) {
-    B("Not initialized");
-    return;
-  }
-  T = !1;
-  const e = t.audioDimensions[0];
-  self.postMessage({
-    type: "start",
-    payload: { totalFrames: e, frameInterval: j }
-  });
-  let n = 0, a = 1;
-  const s = {}, u = _.inputNames[0], d = _.inputNames[1], l = _.outputNames[0];
-  v = "", p && (p.close(), p = null);
-  for (let o = 0; o < e && !T; o++) {
-    let i = null, c = null, r = null, m = null;
-    try {
-      const f = R.images[n];
-      i = K(f.tensor_file), f.full_image !== v && (p && p.close(), p = await Y(f.full_image), v = f.full_image);
-      const h = ae(
-        t.audioFeatures,
-        t.audioDimensions,
-        o
-      );
-      c = new W("float32", h, [
-        1,
-        F,
-        G,
-        V
-      ]), s[u] = i, s[d] = c, r = await _.run(s);
-      const N = r[l];
-      if (!N)
-        throw new Error("Missing inference output tensor");
-      m = N;
-      const U = await te(N, f, p);
-      self.postMessage(
-        { type: "frame", payload: { frame: U, index: o } },
-        [U]
-      );
-      const O = Q(
-        n,
-        R.images.length,
-        a
-      );
-      n = O.nextIndex, a = O.nextDirection;
-    } catch (f) {
-      B(f?.message || "Run Error");
-      break;
-    } finally {
-      i && i.dispose(), c && c.dispose(), m && m.dispose(), i = null, c = null, m = null, r = null;
-    }
-  }
-  p && (p.close(), p = null), v = "", self.postMessage({ type: "done" });
-};
-self.onmessage = (t) => {
-  const e = t.data;
-  e.type === "init" ? se(e) : e.type === "run" ? oe(e) : e.type === "stop" && (T = !0);
-};