GPU-optimized AI, Machine Learning, & HPC Software

NVIDIA

QuartzNet is a Jasper-like network that uses separable convolutions and larger filter sizes. It has comparable accuracy to Jasper while having much fewer parameters. This particular model has 15 blocks each repeated 5 times.

Model

NVIDIA

STT Zh Citrinet 1024 Gamma 0.25

Citrinet-1024 model with kernel scaling factor (gamma) of 25%, which has been trained on the open-source Aishell-2 Mandarin Chinese corpus.

Model

NVIDIA

STT Zh Citrinet 512

Citrinet 512 model trained on Aishell-2 Mandarin corpus

Model

NVIDIA

STT Zh Conformer-Transducer Large

Conformer-Transducer-Large model for Mandarin Automatic Speech Recognition, Trained on Aishell-2 Mandarin Chinese corpus.

Model

NVIDIA

TTS Zh Fastpitch HifiGan SFSpeech

This model card includes two Mandarin Chinese models: 1) FastPitch Mel-spectrogram generator trained on SF Chinese/English Bilingual Speech dataset; 2) HiFiGAN vocoder trained on Mel-spectrograms predicted by the FastPitch.

Model

NVIDIA

NMT Zh En Transformer6x6

Neural Machine Translation (NMT) model to translate from Chinese to English

Model

—

Frame-VAD Multilingual MarbleNet

Frame-based VAD model using MarbleNet model trained with real multilingual data and synthetic English data.

Model

NVIDIA

NMT En Zh Transformer6x6

Neural Machine Translation (NMT) model to translate from English to Chinese

Model