tiny-voca-ja / config.json
gachimoto's picture
v2: fine-tune on pyopenjtalk-plus re-annotated data (step 114,000); 16-sentence eval normalized CER 1.6% -> 1.0%, UTMOS 3.73
0e57796 verified
Raw History Blame Contribute Delete
2.36 kB
{
"model": {
"sample_rate": 24000,
"hop": 256,
"n_fft": 1024,
"n_mels": 80,
"f0_min": 50.0,
"f0_max": 600.0,
"speakers": [
"F2",
"M2"
],
"vocab_size": 84,
"channels": 192,
"heads": 4,
"feedforward": 640,
"kernel": 7,
"encoder_layers": 4,
"prosody_layers": 2,
"acoustic_layers": 5,
"dropout": 0.1,
"predictor_channels": 128,
"synth_fft": 512,
"synth_hop": 128,
"noise_fft": 256,
"noise_hop": 64,
"cep_causal": 80,
"cep_anti": 16,
"noise_cep": 32,
"harmonic_limit": 0.47,
"max_tokens": 512,
"max_frames": 3000,
"mel_mean": -5.0,
"mel_scale": 5.0,
"energy_mean": -4.0,
"energy_scale": 2.0,
"pitch_reference_hz": 200.0
},
"step": 114000,
"sample_rate": 24000,
"hop": 256,
"voices": [
"F2",
"M2"
],
"frontend": {
"name": "pyopenjtalk-plus",
"version": "oj-1",
"mode": "g2p_prosody",
"pyopenjtalk_plus_version": "0.4.1.post10",
"adapter": "oj_frontend.py (frontend-swap)",
"fingerprint": "a03342020f6ecbc6c3d1c7d661ae3e6f3b67f2c657435645ac7a2b87605042d1"
},
"stress": {
"none": 0,
"primary": 1,
"secondary": 2
},
"vocab": {
"<pad>": 0,
"<bos>": 1,
"<eos>": 2,
"<sp>": 3,
"<unk>": 4,
",": 5,
".": 6,
"?": 7,
"!": 8,
";": 9,
":": 10,
"\"": 11,
"'": 12,
"(": 13,
")": 14,
"—": 15,
"…": 16,
"p": 17,
"b": 18,
"t": 19,
"d": 20,
"k": 21,
"ɡ": 22,
"f": 23,
"v": 24,
"θ": 25,
"ð": 26,
"s": 27,
"z": 28,
"ʃ": 29,
"ʒ": 30,
"h": 31,
"m": 32,
"n": 33,
"ŋ": 34,
"l": 35,
"ɹ": 36,
"j": 37,
"w": 38,
"tʃ": 39,
"dʒ": 40,
"ɾ": 41,
"ʔ": 42,
"ə": 43,
"ɐ": 44,
"ɚ": 45,
"ᵻ": 46,
"ɪ": 47,
"i": 48,
"iː": 49,
"ɛ": 50,
"æ": 51,
"ʌ": 52,
"ɑː": 53,
"ɔ": 54,
"ɔː": 55,
"ʊ": 56,
"uː": 57,
"ɜː": 58,
"eɪ": 59,
"aɪ": 60,
"ɔɪ": 61,
"oʊ": 62,
"aʊ": 63,
"iə": 64,
"oː": 65,
"ɑːɹ": 66,
"ɔːɹ": 67,
"oːɹ": 68,
"ɪɹ": 69,
"ɛɹ": 70,
"ʊɹ": 71,
"aɪɚ": 72,
"aɪə": 73,
"əl": 74,
"n̩": 75,
"ɯᵝ": 76,
"ɽ": 77,
"ɕ": 78,
"ɴ": 79,
"tɕ": 80,
"ɸ": 81,
"dʑ": 82,
"kʲ": 83
},
"onnx": {
"file": "onnx/model.onnx",
"inputs": [
"tokens",
"stress",
"voice",
"speed",
"pitch_shift",
"pitch_range"
],
"outputs": [
"audio",
"durations"
]
}
}