| 379 | |
| 380 | |
| 381 | def process_tts_multi(text, save_dir, voice1, voice2): |
| 382 | pattern = r'\(s(\d+)\)\s*(.*?)(?=\s*\(s\d+\)|$)' |
| 383 | matches = re.findall(pattern, text, re.DOTALL) |
| 384 | |
| 385 | s1_sentences = [] |
| 386 | s2_sentences = [] |
| 387 | |
| 388 | pipeline = KPipeline(lang_code='a', repo_id='weights/Kokoro-82M') |
| 389 | for idx, (speaker, content) in enumerate(matches): |
| 390 | if speaker == '1': |
| 391 | voice_tensor = torch.load(voice1, weights_only=True) |
| 392 | generator = pipeline( |
| 393 | content, voice=voice_tensor, # <= change voice here |
| 394 | speed=1, split_pattern=r'\n+' |
| 395 | ) |
| 396 | audios = [] |
| 397 | for i, (gs, ps, audio) in enumerate(generator): |
| 398 | audios.append(audio) |
| 399 | audios = torch.concat(audios, dim=0) |
| 400 | s1_sentences.append(audios) |
| 401 | s2_sentences.append(torch.zeros_like(audios)) |
| 402 | elif speaker == '2': |
| 403 | voice_tensor = torch.load(voice2, weights_only=True) |
| 404 | generator = pipeline( |
| 405 | content, voice=voice_tensor, # <= change voice here |
| 406 | speed=1, split_pattern=r'\n+' |
| 407 | ) |
| 408 | audios = [] |
| 409 | for i, (gs, ps, audio) in enumerate(generator): |
| 410 | audios.append(audio) |
| 411 | audios = torch.concat(audios, dim=0) |
| 412 | s2_sentences.append(audios) |
| 413 | s1_sentences.append(torch.zeros_like(audios)) |
| 414 | |
| 415 | s1_sentences = torch.concat(s1_sentences, dim=0) |
| 416 | s2_sentences = torch.concat(s2_sentences, dim=0) |
| 417 | sum_sentences = s1_sentences + s2_sentences |
| 418 | save_path1 =f'{save_dir}/s1.wav' |
| 419 | save_path2 =f'{save_dir}/s2.wav' |
| 420 | save_path_sum = f'{save_dir}/sum.wav' |
| 421 | sf.write(save_path1, s1_sentences, 24000) # save each audio file |
| 422 | sf.write(save_path2, s2_sentences, 24000) |
| 423 | sf.write(save_path_sum, sum_sentences, 24000) |
| 424 | |
| 425 | s1, _ = librosa.load(save_path1, sr=16000) |
| 426 | s2, _ = librosa.load(save_path2, sr=16000) |
| 427 | # sum, _ = librosa.load(save_path_sum, sr=16000) |
| 428 | return s1, s2, save_path_sum |
| 429 | |
| 430 | def generate(args): |
| 431 | rank = int(os.getenv("RANK", 0)) |