Myduso/AzurLane-Irodori-TTS-LoRA

text-to-speech model by Myduso

Open original source ↗

From the publisher

Model card & documentation

Source preview

Read the publisher’s intended use, setup instructions, evaluations and limitations. The original model card is the source of truth.

仅供学习交流使用,禁止商业用途;角色与语音版权归原作方(碧蓝航线)及各位声优。 基于 Aratako/Irodori-TTS-v4.1-Small 微调的五个碧蓝航线角色声线,每个角色一套 LoRA 适配器加一个说话人嵌入(speaker inversion)。训练所用的官方语音与日文转写见数据集仓库 Myduso/AzurLane-Irodori-TTS-Dataset。 每个目录三个文件: 说话人嵌入的文件名必须以 .speaker.safetensors 结尾,换成别的名字加载器会直接拒绝。 Irodori 是整段生成的模型,上游没有音频流式输出,每段至少要等一次完整合成。我们在桌宠里自己加了一层流式:服务端边收大模型的文字边切句,切出一句就送去合成,几句并行合成,播放端按序号依次播。音色由嵌入固定,每句只换描述,中途不用切换参考音频。 本项目实测(同一段三句话、同一个桌宠聊天框,只换了语音通道): 第一声没有变快,这是每段合成的固定成本;变化在句与句之间,下一句在上一句播放时就已经合成好了。实现和踩坑记录见教程 05 流式语音。 底模必须用 v4.1-Small,adapterconfig.json 里没有记录底模路径,挂到别的底模上不会报错,但生成质量会明显变差。音频 codec(Aratako/Semantic-DACVAE-Japanese-32dim)由 infer.py 自动下载。 --lora-adapter 指向角色目录,--ref-embed 指向该角色的说话人嵌入。两者可以同时用,--ref-embed 不能和 --ref-wav / --ref-wavs / --no-ref 同时出现。 换角色只改两处路径(--lora-adapter 与 --ref-embed)。 大凤额外需要统一引导: 说明:albion 的 LoRA 训练时见过自由描述,可以用任意简短日语描述当 --caption;taihou / shinano / brest / jinshi 训练时用的是一组固定的日语描述,建议从数据集仓库 metadata.jsonl 的 caption 字段里挑原句使用,分布外的描述效果不稳定。--num-steps 40 是我们生产环境的步数;--cfg-scale-caption 是可调的引导强度,按听感自行调整。 Non-commercial use only. Character names and in-game voices belong to the Azur Lane rights holders and the respective voice actors; this repository contains only LoRA adapters and speaker-inversion embeddings fine-tuned on official in-game Japanese voice lines, not the original audio and not the base model. Five Azur Lane voices are provided on top of Aratako/Irodori-TTS-v4.1-Small (MIT). Inference needs no reference audio: the voice comes from the trained speaker embedding .speaker.safetensors, and style and emotion are guided by a short Japanese caption (the upstream documentation calls --caption…

Myduso/AzurLane-Irodori-TTS-LoRA on Hugging Face A short preview, not the full document.

Read the full model card ↗ · Preview checked 2026-11-21T08:46:33.471Z

Inside the original model card — Document outline
  1. AzurLane-Irodori-TTS-LoRA
  2. 角色
  3. 特点
  4. 接进实时聊天:我们的流式做法
  5. 用法
  6. 1. 安装推理代码
  7. 2. 下载底模和本仓库
  8. 3. 挂 LoRA 加说话人嵌入
  9. 4. 合成(日语文本)
  10. 许可与声明
  11. English
  12. 致谢

Headings are captured from the source. Links open the publisher’s document, not a locally hosted copy.

Documentation belongs to its respective authors. Reported project/model license: cc-by-nc-4.0. A listing is not a grant of reuse or training rights. Confirm the document’s own terms at the source.

Model overview

Myduso/AzurLane-Irodori-TTS-LoRA is a text-to-speech model repository published by Myduso on Hugging Face. A library was not reported.

This page summarizes Hub metadata. For intended use, training data, evaluation results and limitations, consult the original model card.

Model facts

Task
text-to-speech
Library
Not reported
Recent downloads (30 days)
0
Cumulative likes
0
Hugging Face trending score
0
Reported safetensors parameters
Not reported
Architecture
Not reported
License
cc-by-nc-4.0
Access
Not gated by Hugging Face
Created
2026-11-21T06:01:12.000Z
Last modified
2026-11-21T06:05:21.000Z

Compatibility and lineage

Reported languages: ja

Reported base models: Aratako/Irodori-TTS-v4.1-Small

Parameter count is not a RAM/VRAM requirement. Check precision, quantization, context length and runtime compatibility in the model card; no hardware or API-cost claim is inferred here.

Use and evaluate

Check license terms and access requirements first. Review the model card’s documented loading instructions, evaluations and safety limitations. Benchmark scores are not imported by this directory, and popularity is not an accuracy ranking.

No model weights are downloaded or executed by AltAPIs. Never enable remote model code without reviewing it.

Model card and usage instructions

Model files and configuration

Community discussion

Source and freshness

Source: Hugging Face Hub. Metadata observed 2026-11-21T07:02:33.996Z. Daily imports are snapshots, not real-time monitoring.

Popularity and source listings do not establish security, suitability, licensing rights or benchmark performance.

Open original source

Related models

hexgrad/Kokoro-82M

audio-cpp/audio.cpp-gguf

coqui/XTTS-v2

k2-fsa/OmniVoice

ResembleAI/chatterbox

pnnbao-ump/VieNeu-TTS-v3-Turbo