VitsInputTemplate.json

[
    {
        "Tokens": "私は誰？",
        "Seq": ["w","a","t","a","s","h","i","w","a","d","a","r","e","?"],
        "Tones": [0,0,0,0,0,0,0,0,0,0,0,0,0],
        "Durations": [2,5,2,5,2,2,5,2,5,2,5,2,5],
        "Language": [0,0,0,0,0,0,0,0,0,0,0,0,0],
        "SpeakerMix": [1,0,0],
        "EmotionPrompt": ["sad", "happy"],
        "NoiseScale": 0.666,
        "LengthScale": 1.1,
        "DurationPredictorNoiseScale": 0.333,
        "FactorDpSdp": 0.6,
        "GateThreshold": 0.777,
        "MaxDecodeStep": 114514,
        "Seed": 1919810,
        "SpeakerId": 2,
        "RestTime": 1.0,
        "PlaceHolderSymbol": "|",
        "LanguageID": "JP",
        "G2PAdditionalInfo": "/[Japanese2]"
    },
    {
        "Tokens": "私は鳴瀬しろは",
        "Seq": "[ph]w|a|t|a|s|h|i|h|a|n|a|r|u|s|e|s|h|i|r|o|h|a",
        "Tones": [],
        "Durations": [],
        "Language": [],
        "SpeakerMix": [],
        "EmotionPrompt": [],
        "NoiseScale": 0.666,
        "LengthScale": 1.0,
        "DurationPredictorNoiseScale": 0.333,
        "FactorDpSdp": 0.6,
        "GateThreshold": 0.777,
        "MaxDecodeStep": 1145147,
        "Seed": 1919810,
        "SpeakerId": 0,
        "RestTime": 1.0,
        "PlaceHolderSymbol": "|",
        "LanguageID": "JP",
        "G2PAdditionalInfo": "/[Japanese2]"
    },
    {
        "Tokens": "私は誰？",
        "Seq": [],
        "Tones": [],
        "Durations": [],
        "Language": [],
        "SpeakerMix": [0.666, 0.233, 0.444],
        "EmotionPrompt": ["sad", "happy"],
        "NoiseScale": 0.666,
        "LengthScale": 1.1,
        "DurationPredictorNoiseScale": 0.333,
        "FactorDpSdp": 0.6,
        "GateThreshold": 0.777,
        "MaxDecodeStep": 1145145,
        "Seed": 19198101,
        "SpeakerId": 1,
        "RestTime": -1.0,
        "PlaceHolderSymbol": "|",
        "LanguageID": "JP",
        "G2PAdditionalInfo": "/[Japanese2]"
    },
    {
        "Tokens": "",
        "Seq": "私は鷗です、くみの名前は？",
        "Tones": [],
        "Durations": [],
        "Language": [],
        "SpeakerMix": [],
        "EmotionPrompt": [],
        "NoiseScale": 0.666,
        "LengthScale": 1.1,
        "DurationPredictorNoiseScale": 0.222,
        "FactorDpSdp": 0.6,
        "GateThreshold": 0.777,
        "MaxDecodeStep": 114514,
        "Seed": 1919810,
        "SpeakerId": 3,
        "RestTime": 1.0,
        "PlaceHolderSymbol": "|",
        "LanguageID": "JP",
        "G2PAdditionalInfo": "/[Japanese2]"
    }
]