| 419 | "voxcpm2": { |
| 420 | "kind": "voxcpm2", |
| 421 | "modes": ["offline", "streaming"], |
| 422 | "cpp_bin": "build/debug/bin/voxcpm2_warm_bench", |
| 423 | "python_script": "tests/voxcpm2/voxcpm2_python_warm_bench.py", |
| 424 | "python_conda_env": "qwen3-tts", |
| 425 | "model": "models/VoxCPM2", |
| 426 | "case_catalog": "tests/voxcpm2/voxcpm2_warm_bench_cases.json", |
| 427 | "default_case_name": "cuda_parity_512char", |
| 428 | "wav_cosine_min": 0.95, |
| 429 | "log_mel_cosine_min": 0.95, |
| 430 | }, |
| 431 | "supertonic": { |
| 432 | "kind": "supertonic", |
| 433 | "modes": ["offline"], |
| 434 | "cpp_bin": "build/debug/bin/supertonic_warm_bench", |
| 435 | "python_script": "tests/supertonic/supertonic_python_warm_bench.py", |
| 436 | "python_conda_env": "qwen3-tts", |
| 437 | "model": "models/supertonic-3", |
| 438 | "case_catalog": "tests/supertonic/supertonic_warm_bench_cases.json", |
| 439 | "default_case_name": "multi_path_voice_lang_steps_speed_chunking", |
| 440 | "default_requests_per_session": 6, |
| 441 | "default_warmup": 0, |
| 442 | "wav_cosine_min": 0.90, |
| 443 | "log_mel_cosine_min": 0.90, |
| 444 | }, |
| 445 | "vibevoice": { |
| 446 | "kind": "vibevoice", |
| 447 | "modes": ["offline"], |
| 448 | "cpp_bin": "build/debug/bin/vibevoice_warm_bench", |
| 449 | "python_script": "tests/vibevoice/vibevoice_python_warm_bench.py", |
| 450 | "python_conda_env": "qwen3-tts", |
| 451 | "model": "models/VibeVoice-1.5B", |
| 452 | "case_catalog": "tests/vibevoice/vibevoice_warm_bench_cases.json", |
| 453 | "default_case_name": "short", |
| 454 | "default_requests_per_session": 1, |
| 455 | "default_warmup": 0, |
| 456 | "wav_cosine_min": 0.90, |
| 457 | "log_mel_cosine_min": 0.90, |
| 458 | "length_ratio_is_diagnostic": True, |
| 459 | }, |
| 460 | "irodori_tts": { |
| 461 | "kind": "irodori_tts", |
| 462 | "modes": ["offline"], |
| 463 | "cpp_bin": "build/debug/bin/irodori_tts_warm_bench", |
| 464 | "python_script": "tests/irodori_tts/irodori_tts_python_warm_bench.py", |
| 465 | "python_conda_env": "qwen3-tts", |
| 466 | "model": "models/Irodori-TTS-500M-v3", |
| 467 | "case_catalog": "tests/irodori_tts/irodori_tts_warm_bench_cases.json", |
| 468 | "default_case_name": "short_emoji", |
| 469 | "default_requests_per_session": 1, |
| 470 | "default_warmup": 0, |
| 471 | "wav_cosine_min": 0.98, |
| 472 | "log_mel_cosine_min": 0.98, |
| 473 | "cpp_session_options": ["irodori_tts.weight_type=f32", "irodori_tts.codec_weight_type=f32"], |
| 474 | }, |
| 475 | "heartmula": { |
| 476 | "kind": "heartmula", |
| 477 | "modes": ["offline"], |
| 478 | "cpp_bin": "build/debug/bin/heartmula_warm_bench", |