Codestin Search App

History

540 lines (490 loc) · 25 KB

Raw

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

224

225

226

227

228

229

230

231

232

233

234

235

236

237

238

239

240

241

242

243

244

245

246

247

248

249

250

251

252

253

254

255

256

257

258

259

260

261

262

263

264

265

266

267

268

269

270

271

272

273

274

275

276

277

278

279

280

281

282

283

284

285

286

287

288

289

290

291

292

293

294

295

296

297

298

299

300

301

302

303

304

305

306

307

308

309

310

311

312

313

314

315

316

317

318

319

320

321

322

323

324

325

326

327

328

329

330

331

332

333

334

335

336

337

338

339

340

341

342

343

344

345

346

347

348

349

350

351

352

353

354

355

356

357

358

359

360

361

362

363

364

365

366

367

368

369

370

371

372

373

374

375

376

377

378

379

380

381

382

383

384

385

386

387

388

389

390

391

392

393

394

395

396

397

398

399

400

401

402

403

404

405

406

407

408

409

410

411

412

413

414

415

416

417

418

419

420

421

422

423

424

425

426

427

428

429

430

431

432

433

434

435

436

437

438

439

440

441

442

443

444

445

446

447

448

449

450

451

452

453

454

455

456

457

458

459

460

461

462

463

464

465

466

467

468

469

470

471

472

473

474

475

476

477

478

479

480

481

482

483

484

485

486

487

488

489

490

491

492

493

494

495

496

497

498

499

500

501

502

503

504

505

506

507

508

509

510

511

512

513

514

515

516

517

518

519

520

521

522

523

524

525

526

527

528

529

530

531

532

533

534

535

536

537

538

539

540

<!DOCTYPE html>

<head>

<title>UniTok-Audio: A Unified Audio Generation Framework</title>

<style>

:root {

--primary-color: #2c3e50;

--secondary-color: #3498db;

--accent-color: #e74c3c;

--bg-light: #f8f9fa;

--text-dark: #2c3e50;

--text-gray: #7f8c8d;

--border: #e0e0e0;

--shadow: 0 0 20px rgba(0, 0, 0, 0.1);

}

body {

font-family: 'Segoe UI', Tahoma, Geneva, Verdana, sans-serif;

line-height: 1.7;

color: var(--text-dark);

background-color: #f5f5f5;

margin: 0;

padding: 0;

}

.container {

max-width: 1200px;

margin: 20px auto;

padding: 30px;

background: white;

border-radius: 12px;

box-shadow: var(--shadow);

position: relative;

}

/* Title & Authors */

.title {

text-align: center;

font-size: 2.4em;

font-weight: 700;

color: var(--primary-color);

margin-bottom: 10px;

letter-spacing: -0.5px;

}

.links {

text-align: center;

margin: 15px 0;

font-size: 1.1em;

}

.links a {

margin: 0 16px;

padding: 8px 16px;

background: var(--secondary-color);

color: white !important;

text-decoration: none;

border-radius: 6px;

font-weight: 600;

transition: all 0.3s ease;

}

.links a:hover {

background: #2980b9;

transform: translateY(-2px);

}

.authors p {

margin: 8px 0;

font-size: 1.1em;

font-weight: 500;

color: var(--text-dark);

}

.institution {

font-size: 1em;

color: var(--text-gray);

text-align: center;

margin: 10px 0 30px;

}

/* Section */

.section {

margin-bottom: 50px;

}

.section-title {

font-size: 1.6em;

color: var(--primary-color);

margin-bottom: 25px;

padding-bottom: 10px;

border-bottom: 2px solid var(--border);

font-weight: 600;

display: inline-block;

}

.abstract-content {

padding: 25px;

background-color: var(--bg-light);

border-left: 4px solid var(--secondary-color);

border-radius: 8px;

font-size: 1.15em;

line-height: 1.8;

color: var(--text-dark);

}

/* Figures */

.figure-container {

display: flex;

flex-direction: column;

align-items: center;

gap: 15px;

margin: 20px 0;

}

.figure {

max-width: 100%;

height: auto;

border-radius: 8px;

box-shadow: 0 4px 12px rgba(0, 0, 0, 0.15);

transition: transform 0.3s ease;

}

.figure:hover {

transform: scale(1.02);

}

.figure-caption {

font-size: 0.95em;

color: var(--text-gray);

text-align: center;

margin-top: 8px;

font-style: italic;

}

/* Audio Tables */

.audio-demos {

width: 100%;

overflow-x: auto;

margin-top: 15px;

}

.audio-table {

width: 100%;

border-collapse: collapse;

background: white;

box-shadow: 0 2px 10px rgba(0, 0, 0, 0.08);

border-radius: 8px;

overflow: hidden;

}

.audio-table th {

background-color: #f1f3f5;

color: var(--primary-color);

font-weight: 600;

padding: 14px 12px;

text-align: center;

font-size: 1.1em;

border-bottom: 2px solid var(--border);

}

.audio-table td {

padding: 14px 12px;

text-align: center;

border-bottom: 1px solid var(--border);

vertical-align: middle;

}

.audio-table tr:hover {

background-color: #f8f9ff;

}

audio {

width: 250px;

height: 40px;

border-radius: 8px;

box-shadow: 0 1px 5px rgba(0, 0, 0, 0.1);

}

.audio-name {

font-weight: 600;

color: var(--text-dark);

text-align: left;

padding: 14px;

background: #fafafa;

}

/* Responsive */

@media (max-width: 768px) {

.container {

padding: 15px;

margin: 10px;

}

.title {

font-size: 2em;

}

.links a {

display: block;

margin: 10px auto;

width: fit-content;

}

audio {

width: 100%;

}

.audio-table {

font-size: 0.95em;

}

.abstract-content {

font-size: 1em;

padding: 18px;

}

</style>

</head>

<body>

UniTok-Audio: A Unified Audio Generation Framework Via Universal Discrete Token

</h1>

<a href="#">Chengwei Liu†</a><sup>1</sup>,</span>

<a href="#">Haoyin Yan†</a><sup>1,2</sup>,</span>

<a href="#">Shaofei Xue*</a><sup>1,2</sup>,</span>

<a href="#">Xiaotao Liang</a><sup>1</sup>,</span>

<a href="#">Yinghao Liu</a><sup>1,2</sup>

<a href="#">Zheng Xue</a><sup>1</sup>,</span>

<a href="#">Boyang Zhou</a><sup>1,3</sup>,</span>

</span>

</div>

<span class="author-block"><sup>1</sup>Intelligent Connectivity, Alibaba Group</span>,

<span class="author-block"><sup>2</sup>Tongyi Lab, Alibaba Group</span>

<span class="author-block"><sup>3</sup>Zhejiang University</span>

</div>

<em>†Equal Contribution: Haoyin Yan, Chengwei Liu</em>,

<em>*Corresponding author: Shaofei Xue</em>

</div>

<a href="https://arxiv.org/abs/2510.26372"

class="external-link button is-normal is-rounded is-dark">

</span>

<span>arXiv</span>

</a>

</span>

<a href="https://github.com/alibaba/unified-audio"

class="external-link button is-normal is-rounded is-dark" target="_blank">

</span>

</a>

</span>

</div>

</section>

<h2 class="section-title">Abstract</h2>

Generative modeling has recently achieved remarkable success across text, image, and audio domains, demonstrating powerful capabilities for unified representation learning. However, audio generation models still face challenges in terms of audio quality and generalization ability across tasks. To address these issues, we propose UniTok-Audio, a scalable and extensible framework for unified audio generation tasks. Specifically, 1) UniTok-Audio extracts continuous feature of conditions to generates discrete tokens of target auido in an autoregressive manner; 2) a special task identifier token unifies different learning patterns of multiple tasks in an unified framework; 3) a dual-stream audio codec involving acoustic and semantic branch is developed for high-fidelity waveform reconstruction. Experimental results demonstrate that UniTok-Audio achieves competitive performance in comparation with state-of-the-art task-specific or multi-task systems across five time-aligned tasks: speech restoration, target speaker extraction, speech separation, voice conversion, and language-queried audio source separation. To foster future research, we will open-source our codebase.

</div>

<h2 class="section-title">UniTok-Audio</h2>

<p class="figure-caption">Fig. 1: Overall architecture of UniTok-Audio, where the H-Codec Encoder is only used to generate label tokens during training and excluded during inference.</p>

</div>

UniTok-Audio is a unified discrete-token-based audio generation framework that consists of four parts: (1) a WavLM or HuBERT audio encoder, (2) a CLAP text encoder, (3) a LLaMA-based language model, and (4) a novel H-Codec decoder. The WavLM and HuBERT encoders extract continuous speech features from audio. The LLaMA-based LM takes these features as input and predicts discrete speech tokens generated by H-Codec in an autoregressive manner. Finally, the H-Codec decoder reconstructs enhanced speech from predicted tokens.

Guided by task identifiers and timestep embeddings, the LM processes latent sequences using cross-attention to generate task-specific outputs.

</p>

</div>

<h2 class="section-title">H-Codec</h2>

<p class="figure-caption">Fig. 2: Overall architecture of H-Codec.</p>

</div>

H-Codec features two encoding streams: a self-supervised learning (SSL) stream and a waveform stream. The SSL stream captures semantic-rich information and injects it into the first-layer codec tokens via direct encoding from HuBERT/WavLM features. The waveform stream uses a proven DAC-like framework to encode and decode high-quality audio. Both streams are downsampled to achieve a low frame rate of 25 Hz.

During training, both streams are used to obtain target tokens. During inference, only the decoder is active to generate high-fidelity audio.

</p>

</div>

<h2 class="section-title">H-Codec Result</h2>

<p class="figure-caption">Fig. 3: Speech Reconstruction and Semantic Performance.</p>

</div>

H-Codec achieves the best performance at a token rate of 50 for most metrics. Moreover, its UTMOS score closely matches that of the ground truth, indicating that the reconstructed audio faithfully preserves the original speech quality. We also observe that certain models exceed the ground truth in UTMOS when operating at low token rates. We suspect this occurs because, under limited token constraints, the decoder behaves partly as a generative model—yielding plausible speech output but the alignment with the input was less precise..

</p>

</div>

<h2 class="section-title">Speech Restoration (SR)</h2>

<thead>

<tr>

<th>Clean</th>

<th>Noisy</th>

<th>Enhanced</th>

</tr>

</thead>

<tbody>

<tr>

</tr>

<tr>

</tr>

<tr>

</tr>

<tr>

</tr>

</tbody>

</table>

</div>

<h2 class="section-title">Target Speech Extraction (TSE)</h2>

<thead>

<tr>

<th>Target</th>

<th>Mixture</th>

<th>Enhanced</th>

</tr>

</thead>

<tbody>

<tr>

</tr>

<tr>

</tr>

<tr>

</tr>

<tr>

</tr>

</tbody>

</table>

</div>

<h2 class="section-title">Speech Separation (SS)</h2>

<thead>

<tr>

<th>Mixture</th>

<th>Speaker 1</th>

<th>Speaker 2</th>

</tr>

</thead>

<tbody>

<tr>

</tr>

<tr>

</tr>

<tr>

</tr>

<tr>

</tr>

</tbody>

</table>

</div>

<h2 class="section-title">Voice Conversion (VC)</h2>

<thead>

<tr>

<th>Source</th>

<th>Reference</th>

<th>Converted</th>

</tr>

</thead>

<tbody>

<tr>

</tr>

<tr>

</tr>

<tr>

</tr>

<tr>

</tr>

</tbody>

</table>

</div>

<h2 class="section-title">Language-Queried Audio Source Separation (LASS)</h2>

<thead>

<tr>

<th>Text Query</th>

<th>Source</th>

<th>Converted</th>

</tr>

</thead>

<tbody>

<tr>

<td>the crowd is cheering and giving applause</td>

</tr>

<tr>

<td>someone is beating the drum continuously</td>

</tr>

<tr>

<td>Someone is typing on a keyboard</td>

</tr>

<tr>

<td>a person is pressing the shutter button of the camera to check something</td>

</tr>

</tbody>

</table>

</div>

</body>

</html>

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

FilesExpand file tree

index.html

Latest commit

History

index.html

File metadata and controls