wyh2000

## What? We add voice conversion task in espnet2, and support both parallel data and non-parallel data. ## Why? To support more voice conversion models and datasets ## See also...

Reconstruct image with only z_{sem}, with x_T is sampled from N (0, I)

2

Hi, thanks for sharing this nice work. Could you share some example code for how to reconstruct images by DiffAE when only z_{sem} is encoded from original images but x_T...

We add a HifiGAN based vocoder which can decode hubert tokens to waveform. It supports LJSpeech now. It is aligned with configurations in discrete TTS from [ESPnet](https://github.com/espnet/espnet/pull/5626).

add ASR evaluation for long speech

5

## What? Add evaluation scripts for long speech ASR. ## Why? For long speech (e.g. longer than 20 seconds), we should first split it into shorter segments and then evaluate...

PR SpeechComposer

5

## What? Adding recipe for training SpeechComposer on voice conversion and speech enhancement. ## Why? Allow training for language models on new tasks and adding recipe. ## See also This...

Implementation of EVA: Robust Audiovisual Speech Recognition Models with Mixture-of-Experts

3

## What? This is an implementation of EVA: Robust Audiovisual Speech Recognition Models with Mixture-of-Experts. It supports audiovisual ASR for unconstrained videos. EVA implementation is based on OWSM v3.1, with...

Development plan for ESPnet2 voice conversion

4

We want to develop the ESPnet2 for the voice conversion task. We aim to develop voice conversion systems which support both parallel data and un-parallel data. # Progress & Plan...

Support more than one rejected responses

3

Hi, I'd like to know if DPOTrainer supports more than one rejected sample. As the original DPO paper mentioned, they can support Plackett-Luce model for a set of possible responses.

Training code release

1

Hi, Thanks for your great job! Do you have any plan to release the training code of Video-Salmonn?