---
license: apache-2.0
language:
- en
base_model:
- facebook/wav2vec2-large-xlsr-53
tags:
- phone
- speech
- recognition
- british
---

# GBPhone: British English Phone Recognizer

GBPhone is a phone recognizer trained for British English and producing [SAMPA](https://en.wikipedia.org/wiki/SAMPA_chart_for_English) phone symbols.

GBPhone was fine tuned from the [wav2vec2 XLSR](https://huggingface.co/docs/transformers/en/model_doc/xlsr_wav2vec2) model using a British English dataset.

An example Python script is included. Output is a CSV file with log likelihoods per phone per frame. 
Because the model is trained by CTC, each phone is marked only at the start of each segment, and the blank symbol (blk) is used to pad the rest of the segment.

An example R script is included to display the recognition results.

Mark Huckvale
March 2025