babysor / babysor/MockingBird

【长期】如何改模型,搞出更好效果

Open
#436 2 comments 0 reactions 0 assignees View on GitHub
discussion
Dominant language
Python
Stars
36.9k
Forks
5.2k
PR merge metrics
No merged PRs in 30d

Description

### 合成器Synthesizer
将 synthesizer部分换为 tacotron2 详见 https://github.com/babysor/MockingBird/pull/296
### 声码器
@[AyahaShirane](https://github.com/AyahaShirane)
我发现合成器在达到一定数量的训练后,对质量影响更大的反而是声码器,hifigan能用更好的效果但始终有电噪音,现在有些新的项目采样lpcnet而非wavernn,据说是复杂度要远低于wavernn,但质量优于wavernn,可以达到类似hifigan的效果而不带电噪音,请问有考虑过引入lpcnet之类的新声码器的计划呢?

> 作者却苦于近期精力限制只能势单力薄处理一些小的bug,也看到issue区有不少爱好与开发者想要学习或二次改造更好满足自己需求,不过比较零碎难以展开。为了让项目和AI持续可以给大家提供更多价值,共同学习,我在issue区根据不同主题创建长期交流频道,若留言人数超过20也将建立对应交流群。
> - 如何改参数,搞出更逼真的克隆效果 435
> - 如何改模型,搞出更好效果 436
> - 训练克隆特定人声音&finetune 437
> - 学术/论文讨论/训练分析 438
> - 跨语言支持 440
> - 工程化/新场景讨论(绝不做恶 & 合法合规) 439

Contributor guide

No contributing guide indexed for this repository

Research direction

Start with the Synthesizer section and the linked pull request #296, then review the vocoder discussion covering HiFi-GAN, WaveRNN, and LPCNet. Done would require a decided scope for the model or vocoder change, rather than the current long-term discussion.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
audio-video-rtc, machine-learning
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
20/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.