MooseNet: A Trainable Metric for Synthesized Speech with a PLDA Module
Identifikátory výsledku
Kód výsledku v IS VaVaI
<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216208%3A11320%2F23%3A10475701" target="_blank" >RIV/00216208:11320/23:10475701 - isvavai.cz</a>
Výsledek na webu
<a href="http://dx.doi.org/10.21437/SSW.2023-8" target="_blank" >http://dx.doi.org/10.21437/SSW.2023-8</a>
DOI - Digital Object Identifier
<a href="http://dx.doi.org/10.21437/SSW.2023-8" target="_blank" >10.21437/SSW.2023-8</a>
Alternativní jazyky
Jazyk výsledku
angličtina
Název v původním jazyce
MooseNet: A Trainable Metric for Synthesized Speech with a PLDA Module
Popis výsledku v původním jazyce
We present MooseNet, a trainable speech metric that predicts the listeners' Mean Opinion Score (MOS). We propose a novel approach where the Probabilistic Linear Discriminative Analysis (PLDA) generative model is used on top of an embedding obtained from a self-supervised learning (SSL) neural network (NN) model. We show that PLDA works well with a non-finetuned SSL model when trained only on 136 utterances (ca. one minute training time) and that PLDA consistently improves various neural MOS prediction models, even stateof-the-art models with task-specific fine-tuning. Our ablation study shows PLDA training superiority over SSL model finetuning in a low-resource scenario. We also improve SSL model fine-tuning using a convenient optimizer choice and additional contrastive and multi-task training objectives. The fine-tuned MooseNet NN with the PLDA module achieves the best results, surpassing the SSL baseline on the VoiceMOS Challenge data.
Název v anglickém jazyce
MooseNet: A Trainable Metric for Synthesized Speech with a PLDA Module
Popis výsledku anglicky
We present MooseNet, a trainable speech metric that predicts the listeners' Mean Opinion Score (MOS). We propose a novel approach where the Probabilistic Linear Discriminative Analysis (PLDA) generative model is used on top of an embedding obtained from a self-supervised learning (SSL) neural network (NN) model. We show that PLDA works well with a non-finetuned SSL model when trained only on 136 utterances (ca. one minute training time) and that PLDA consistently improves various neural MOS prediction models, even stateof-the-art models with task-specific fine-tuning. Our ablation study shows PLDA training superiority over SSL model finetuning in a low-resource scenario. We also improve SSL model fine-tuning using a convenient optimizer choice and additional contrastive and multi-task training objectives. The fine-tuned MooseNet NN with the PLDA module achieves the best results, surpassing the SSL baseline on the VoiceMOS Challenge data.
Klasifikace
Druh
O - Ostatní výsledky
CEP obor
—
OECD FORD obor
10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)
Návaznosti výsledku
Projekt
—
Návaznosti
S - Specificky vyzkum na vysokych skolach
Ostatní
Rok uplatnění
2023
Kód důvěrnosti údajů
S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů