UmaVoice_DDSP_6.3 README

项目介绍 / Project Introduction

UmaVoice_DDSP_6.3 是基于DDSP（Differentiable Digital Signal Processing）架构开发的赛马娘主题语音合成模型，核心目标为精准还原赛马娘系列角色的专属音色，可适配语音合成、音色转换等各类音频相关任务。本模型1.0版本现已正式发布多轮训练权重，仅供非商业性质的研究、教育及个人使用。

UmaVoice_DDSP_6.3 is a Uma Musume-themed text-to-speech model developed based on the DDSP (Differentiable Digital Signal Processing) architecture. Its core goal is to accurately restore the exclusive timbre of characters in the Uma Musume series, and it can be applied to various audio-related tasks such as text-to-speech and voice conversion. The 1.0 version of this model has officially released multiple rounds of training weights, which are only for non-commercial research, education and personal use.

训练信息 / Training Information

本模型1.0版本总计完成500轮训练，每轮训练步数约为478步；为便于统一核算与参考，后续相关步数计算均按每轮500步进行近似处理，各轮次训练权重对应的具体步数如下：

The 1.0 version of this model has completed a total of 500 rounds of training, with approximately 478 steps per round. For the convenience of unified calculation and reference, the subsequent calculation of relevant steps is approximated at 500 steps per round. The specific number of steps corresponding to the training weights of each round is as follows:

• 100轮：约50000步 / 100 rounds: approximately 50,000 steps

• 140轮：约70000步 / 140 rounds: approximately 70,000 steps

• 160轮：约80000步（推荐）/ 160 rounds: approximately 80,000 steps (recommended)

• 200轮：约100000步 / 200 rounds: approximately 100,000 steps

• 250轮：约124000步 / 250 rounds: approximately 124,000 steps

• 300轮：约150000步 / 300 rounds: approximately 150,000 steps

• 350轮：约174000步 / 350 rounds: approximately 174,000 steps

• 400轮：约200000步 / 400 rounds: approximately 200,000 steps

• 最终产品：238000步 / Final Product: approximately 238,000 steps

包含人物角色 / Included Characters

本模型涵盖以下赛马娘角色的专属音色，覆盖多种声线风格，可满足不同场景下的使用需求：

This model covers the exclusive timbres of the following Uma Musume characters, including a variety of voice styles, which can meet the usage needs in different scenarios:

说明：模型说话人选择界面采用中文显示，若使用者对中文不熟悉，请参考下方对照表格。

Note: The selection interface of the model speaker is displayed in Chinese. If the user is not familiar with Chinese, please refer to the comparison table below.

中文角色名 / Chinese Character Name	英文角色名 / English Character Name
东商变革	Tosho Revolution
东海帝皇	Tokai Teio
东瀛佐敦	Eishin Jordan
中山庆典	Nakayama Festa
丸善斯基	Maruzensky
也文摄辉	Yamanin Zephyr
伏特加	Vodka
优秀素质	Nice Nature
八重无敌	Yaeno Muteki
北港火山	Hokko Tarumae
北部玄驹	Kitasan Black
千明代表	Mr. C.B.
双涡轮	Twin Turbo
吉兆	Kichijo
名将怒涛	Meisho Doto
大和赤骥	Daiwa Scarlet
大拓太阳神	Daitaku Helios
大树快车	Taiki Shuttle
奇锐骏	Cheval Grand
好歌剧	T.M. Opera O
富士奇迹	Fuji Kiseki
小林历奇	Copano Rickey
小栗帽	Oguri Cap
川上公主	Kawakami Princess
帝王光辉	King Halo
待兼福来	Matikanefukukitaru
成田大进	Narita Taishin
成田白仁	Narita Brian
摩耶重炮	Mayano Top Gun
新光风	Shinko Windy
新宇宙	Neo Universe
无声铃鹿	Silence Suzuka
春乌拉拉	Haru Urara
曼城茶座	Manhattan Cafe
樱花千代王	Sakura Chiyono O
樱花桂冠	Sakura Laurel
樱花进王	Sakura Bakushin O
气槽	Air Groove
爱丽数码	Agnes Digital
爱丽速子	Agnes Tachyon
爱慕织姬	Admire Vega
特别周	Special Week
玉藻十字	Tamamo Cross
琵琶晨光	Biwa Hayahide
目白光明	Mejiro Bright
目白善信	Mejiro Palmer
目白多伯	Mejiro Dober
目白赖恩	Mejiro Ryan
目白阿尔丹	Mejiro Ardan
目白麦昆	Mejiro McQueen
真弓快车	Curren Chan
真机伶	Smart Falcon
神鹰	El Condor Pasa
稻荷一	Inari One
空中神宫	Air Shakur
第一红宝石	Daiichi Ruby
米浴	Rice Shower
美妙姿势	Marvelous Sunday
美浦波旁	Mihono Bourbon
胜利奖券	Winning Ticket
艾尼斯风神	Ines Fujin
草上飞	Grass Wonder
荒漠英雄	Admire Desert
荣进闪耀	Eishin Flash
菱亚马逊	Hishi Amazon
菱曙	Hishi Akebono
菱钻奇宝	Satono Diamond
西野花	Nishino Flower
诗歌剧	Seeking the Pearl
谷水琴蕾	Tanino Gimlet
超级小海湾	Super Creek
醒目飞鹰	Smart Falcon
采珠	Seeking the Pearl
里见光钻	Satono Diamond
雪之美人	Yukino Bijin
青云天空	Seiun Sky
青竹回忆	Bamboo Memory
鲁道夫象征	Symboli Rudolf
黄金城	Gold City
黄金船	Gold Ship

特别发布 / Special Release

本项目单独发布160轮（按每轮500步核算，约80000步）权重对应的SVC Fusion模型包，该模型包的具体使用方法，请参考对应链接内的相关说明，链接地址：SVC Fusion 模型包链接。

This project separately releases the SVC Fusion model package corresponding to the weights of 160 rounds (calculated at 500 steps per round, approximately 80,000 steps). For the specific usage method of this model package, please refer to the relevant instructions in the corresponding link, link address:SVC Fusion model package link.

开源协议 / Open Source License

本项目严格遵循 CC-BY-NC-SA-4.0（知识共享 署名-非商业性使用-相同方式共享 4.0 国际版）开源协议，具体条款如下：

This project strictly abides by the CC-BY-NC-SA-4.0 (Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International) open source license, and the specific terms are as follows:

• 署名（BY）：任何单位或个人在使用、分发、修改本模型及相关权重时，必须完整保留原作者信息及项目来源，不得擅自篡改或省略相关信息。 / Attribution (BY): Any unit or individual must fully retain the original author information and project source when using, distributing or modifying this model and related weights, and shall not arbitrarily tamper with or omit relevant information.

• 非商业（NC）：严禁任何形式的商业使用行为，包括但不限于将本模型用于出租、售卖、盈利性服务、商业衍生开发等，一经发现，将追究相关责任。 / NonCommercial (NC): Any form of commercial use is strictly prohibited, including but not limited to using this model for rental, sale, profitable services, commercial derivative development, etc. Any violation will be held accountable.

• 相同方式共享（SA）：若对本模型或其权重进行修改、二次创作等衍生开发，所产生的衍生作品必须采用与本协议完全一致的CC-BY-NC-SA-4.0协议进行发布，不得更改协议条款。 / ShareAlike (SA): If this model or its weights are modified, recreated or otherwise developed derivatively, the resulting derivative works must be released under the CC-BY-NC-SA-4.0 license that is completely consistent with this agreement, and the terms of the agreement shall not be changed.

• 免责声明：本模型按“现状”提供，作者不对模型的准确性、完整性、适用性作出任何明示或暗示的担保，亦不承担任何因使用、分发、修改本模型及相关权重所导致的法律责任、经济损失或其他相关风险。 / Disclaimer: This model is provided "as is". The author does not make any express or implied guarantees regarding the accuracy, completeness and applicability of the model, nor does he bear any legal liability, economic loss or other related risks caused by the use, distribution or modification of this model and related weights.

使用范围 / Scope of Use

1. 教育教学与个人收听：允许在正规教育教学活动中使用本模型，允许以个人收听为唯一目的的非商业性使用，不得用于其他非授权场景。 / 1. Education, teaching and personal listening: The use of this model in formal education and teaching activities is permitted, and non-commercial use for personal listening as the sole purpose is permitted, and it shall not be used in other unauthorized scenarios.

2. 网络视频/音频平台使用：允许利用本模型制作相关作品并在网络视频、音频平台发布，但严禁开启任何形式的商业化行为，包括但不限于收费、订阅、植入广告等；同时，必须在作品的借物表（或致谢名单）中清晰、准确注明本模型的完整名称及项目地址。 / 2. Use on online video/audio platforms: It is permitted to use this model to create related works and publish them on online video and audio platforms, but any form of commercial behavior is strictly prohibited, including but not limited to charging, subscribing, inserting advertisements, etc.; at the same time, the complete name and project address of this model must be clearly and accurately indicated in the credit list (or acknowledgment list) of the work.

3. 研究与二次开发：允许科研机构、高校及个人将本模型用于非商业性质的研究活动及二次开发，衍生作品需严格遵循本项目开源协议要求。 / 3. Research and secondary development: Scientific research institutions, universities and individuals are permitted to use this model for non-commercial research activities and secondary development, and derivative works must strictly comply with the requirements of the open source license of this project.

使用说明 / Usage Instructions

1. 本模型的使用范围严格限定为非商业、研究及教育目的，严禁任何形式的商业用途，违反者需承担相应法律责任； / 1. The scope of use of this model is strictly limited to non-commercial, research and educational purposes, and any form of commercial use is strictly prohibited. Violators shall bear corresponding legal responsibilities;

2. 使用本模型时，需严格遵守CC-BY-NC-SA-4.0协议的所有条款，完整保留原作者信息及项目来源，不得擅自违反协议要求； / 2. When using this model, all terms of the CC-BY-NC-SA-4.0 license must be strictly followed, the original author information and project source must be fully retained, and the requirements of the license shall not be arbitrarily violated;

3. 不同轮次的训练权重对应不同的模型训练程度，使用者可根据自身实际使用需求，测试并选择适配性最佳的权重版本； / 3. Different rounds of training weights correspond to different model training levels. Users can test and select the most suitable weight version according to their actual usage needs;

4. SVC Fusion模型包可通过上述指定超链接获取，其具体使用方法、操作规范及注意事项，請参考链接内的相关说明文档； / 4. The SVC Fusion model package can be obtained through the above-specified hyperlink. For its specific usage methods, operation specifications and precautions, please refer to the relevant instruction documents in the link;

5. 需特别说明：推荐权重仅为通用性参考，并不一定适配所有使用情景，建议使用者测试多个权重版本以获得最佳效果；个人使用本模型生成的任何内容，均与本模型作者无任何关联，相关内容的合法性、合规性及产生的一切责任，均由使用者自行承担。 / 5. It should be specially noted that the recommended weights are only for general reference and may not be suitable for all usage scenarios. It is recommended that users test multiple weight versions to achieve the best effect; any content generated by individuals using this model is not related to the author of this model, and the legality, compliance and all responsibilities arising from the relevant content shall be borne by the user.