Non-streaming ASR

This page describes how to use the C++ frontend of sherpa for non-streaming ASR.

We use pretrained models from the following two datasets for demonstration: GigaSpeech and WenetSpeech.

The pretrained model from GigaSpeech uses BPE as modeling units, while the model from WenetSpeech uses Chinese characters.

Demo with pretrained models

• C++ APIs
• Pretrained model with GigaSpeech
  • Download the pretrained model
  • Decode a single wave
  • Decode multiple waves in parallel
  • Decode wav.scp
  • Decode feats.scp
• Pretrained model with WenetSpeech
  • Download the pretrained model
  • Decode a single wave
  • Decode multiple waves in parallel
  • Decode wav.scp
  • Decode feats.scp