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

[Question] Pitch during training

Open
#1,083 2 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

following up
Dominant language
Python
Stars
38.4k
Forks
5.3k
PR merge metrics
No merged PRs in 30d

Description

Hello,

I am looking to modify the model a bit and had a few questions if anyone can shed some light on it :)

Why is the pitch added to the text encoder (enc_p) as well as the GeneratorNSF (dec) during training?
Wouldn't only the GeneratorNSF be what creates the output sound and therefore need the pitch to guide it?

I'm also curious why the text encoder is used in inference, according to the VITS paper it uses the posterior encoder (enc_q) in voice conversion rather than the text encoder (prior encoder/enc_p). Any insight here would be very helpful :D
image

Many thanks in advance!

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

Start by tracing the training and inference paths around the text encoder (enc_p), GeneratorNSF (dec), and posterior encoder (enc_q). Compare their roles with the VITS paper and determine whether the issue can be answered from the current model flow; done means providing a clear explanation of pitch conditioning and inference encoder selection.

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
15/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.