Home Console Docs News Papers Tutorials Datasets Wiki SOTA LLM Models GPU Leaderboard Events

English

Speech Synthesis On Libritts

Metrics

M-STFT

MCD

PESQ

Periodicity

V/UV F1

Results

Performance results of various models on this benchmark

						Paper Title	Repository
PeriodWave-Turbo-L	0.7358	-	4.454	0.0528	0.9756	Accelerating High-Fidelity Waveform Generation via Adversarial Flow Matching Optimization
BigVGAN-v2	0.7026	0.2903	4.362	0.0593	0.9793	BigVGAN: A Universal Neural Vocoder with Large-Scale Training
EVA-GAN-big	0.7982	-	4.3536	0.0751	0.9745	EVA-GAN: Enhanced Various Audio Generation via Scalable Generative Adversarial Networks
PeriodWave + FreeU	1.0269	-	4.248	0.0765	0.9651	PeriodWave: Multi-Period Flow Matching for High-Fidelity Waveform Generation
RFWave	-	-	4.228	0.090	0.968	RFWave: Multi-band Rectified Flow for Audio Waveform Reconstruction
BigVSAN (w/ snakebeta)	0.7992	0.4129	4.120	0.0924	0.9644	BigVSAN: Enhancing GAN-based Neural Vocoders with Slicing Adversarial Network
BigVSAN	0.7881	0.3381	4.116	0.0935	0.9635	BigVSAN: Enhancing GAN-based Neural Vocoders with Slicing Adversarial Network
EVA-GAN-base	0.9485	-	4.0330	0.0942	0.9658	EVA-GAN: Enhanced Various Audio Generation via Scalable Generative Adversarial Networks
BigVGAN	0.7997	0.3745	4.027	0.1018	0.9598	BigVGAN: A Universal Neural Vocoder with Large-Scale Training
Vocos	-	-	3.70	0.101	0.9582	Vocos: Closing the gap between time-domain and Fourier-based neural vocoders for high-quality audio synthesis
BigVGAN-base	0.8788	0.4564	3.519	0.1287	0.9459	BigVGAN: A Universal Neural Vocoder with Large-Scale Training
WaveGlow	1.3099	2.3591	3.138	0.1485	0.9378	WaveGlow: A Flow-based Generative Network for Speech Synthesis
WaveFlow	1.1120	1.2455	3.027	0.1416	0.9410	WaveFlow: A Compact Flow-based Model for Raw Audio
HiFi-GAN	1.0017	0.6603	2.947	0.1565	0.9300	HiFi-GAN: Generative Adversarial Networks for Efficient and High Fidelity Speech Synthesis
SC-WaveRNN	2.2358	1.8854	1.701	0.3044	0.8144	Speaker Conditional WaveRNN: Towards Universal Neural Vocoder for Unseen Speaker and Recording Conditions

0 of 15 row(s) selected.