Phoneme and articulatory analysis framework for Kaldi/ESPNet ASR

This works is a phoneme and articulatory analysis framework based on lexicon-based grapheme to phoneme mappings. Phoneme recognition is known to be difficult, as the task is highly contextual, nevertheless, phoneme analysis should be a first step in order to analyse where understanding of ASR models are lacking. This approach can use decoded sentences from word-level ASRs to quantify their performing on phonemes and articulatory features.

The code in this repository can be used to calculate PER and AFER based on a Kaldi wer details file.

What does it do

• Implements a new variant of PER on word-level ASR
• Implements a new error rate, the AFER (Articulatory Feature Error Rate) on word-level ASR

Requirements

pip install git+https://github.com/karkirowle/benzaiten_asr.git

Example for PER extraction

See example.py

Languages supported

🇺🇸 🇳🇱

Future plans

• Support for multiple languages, and g2p models in PER
• Fully support MoA-PoA for Dutch/Mandarin phonemes