Hugging Face's logo

vitouphy
/
wav2vec2-xls-r-300m-timit-phoneme

Automatic Speech Recognition
Transformers
PyTorch
Safetensors
English
wav2vec2
Generated from Trainer
Eval Results
Inference Endpoints
Model card Files Community
8

dx phoneme

#1
by kalbin - opened
Discussion
kalbin
•
edited Jun 21, 2022

I fed in SX189.wav from TIMIT and got back "dɪs t rɔɪ ɛv rifaɪl rəleɪdxɪ tumaɪ ɑdxɪ s"

I think the /dx/ phoneme is still in ARPAbet form instead of IPA in vocab.json

kalbin

I think it should be replaced with [ɾ]

vitouphy
 Owner
•
edited Jun 21, 2022

Hi @kalbin , nice observation. thank you for pointing that out. I will look into it.

kalbin

No problem!
here's my source btw: https://github.com/kosuke-kitahara/xlsr-wav2vec2-phoneme-recognition/blob/main/Fine_tuning_XLSR_Wav2Vec2_for_Phoneme_Recognition.ipynb

vitouphy
 Owner

Change the /dx/ ARPAbet into phoneme [ɾ].

vitouphy changed discussion status to closed

Sign up or log in to comment