asr-rescoring

Rescoring methods for end-to-end Automatic Speech Recognition. The python scripts in this repository cover the following features:

• Fine-tune a GPT-2 language model given a custom dataset
• Run speech recognition on a given WAV file and get the list of transcription candidates
• Score a list of phrases (rescoring) with a language model (e.g. the fine-tuned GPT2)

See the Medium article for more details:

https://medium.com/@diegofustesfic/asr-rescoring-with-fine-tuned-gpt-70f7f6323565