NeuralSP

When it fits

Use NeuralSP when you need an implementation-oriented starting point for end-to-end ASR or language-model experiments and can work with its Python/PyTorch code and Kaldi-related setup.

Who it is for

• Researchers comparing end-to-end ASR architectures and decoding approaches.
• Developers training speech-recognition or language models on the corpora named by the project.
• Practitioners exploring streaming-oriented recognition mechanisms or language-model fusion.

Documented uses

• Prototype ASR systems using CTC, RNN-transducer, or attention-based decoding.
• Compare RNN, Transformer, Conformer, and convolutional encoder designs.
• Train or evaluate models using supported speech and text corpora.
• Experiment with recurrent, convolutional, or Transformer-based language models.