RVC-Project / RVC-Project/Retrieval-based-Voice-Conversion-WebUI

带音高的特征集实验

Open
#697 9 comments 7 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

good first issue
Dominant language
Python
Stars
38.4k
Forks
5.3k
PR merge metrics
No merged PRs in 30d

Description

开发者们好,我想介绍一下我在我的fork(https://github.com/matthew99a/Retrieval-based-Voice-Conversion-WebUI)上做的一个实验。

我用自己很喜欢的一个偏美声歌手训练的模型,最开始听起来不是很像。然后我发现了一个现象:这个歌声在发同一个元音的时候,高音的发音方式和低音的发音方式会不同(比如高音的ah(啊)听感在ah和oo(乌)之间),而HuBERT模型是不能准确的把它们识别成同一种特征的。我因此意识到模型不准确的原因可能是因为模型里训练的特征是不考虑音高的。这样的实现方式在匹配的时候,高音的特征可能与训练集里的低音特征匹配。由于美声歌手往往不会用这个特征唱高音,因此生成出来的高音也会很奇怪。

在我的实验中,我把特征加了一维(v2版本的768维变769维)表示音高,从而使得匹配的时候能够优先匹配训练集里音高接近的特征。使用两个小时的训练集重新训练特征之后,在我的实验中效果好了一大截,当然代价是咬字略微差了一些。当然我也清楚这样做的坏处是训练集要求太高,因为如果训练集太小,可以用来匹配的音高接近的特征就很少,对咬字的牺牲就会很大,并且也许这种实现只有对美声歌手等特殊的唱法非常有效果(不是很清楚一般的流行唱法的高音和低音HuBERT特征差距多大)。

希望这个实验报告能起到一些抛砖引玉的作用。

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

The report describes a fork-based experiment that changes v2 features from 768 to 769 dimensions by adding pitch. Start by reviewing the fork and the repository's feature extraction and matching paths; compare the pitch-aware result with the original on singing data. Done would require a defined implementation scope and evidence that the quality tradeoff is acceptable.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
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.