espnet/mediaspeech-fr-hubert

Usage

import librosa
from espnet2.bin.asr_inference import Speech2Text

speech2text = Speech2Text.from_pretrained(model_tag="espnet/mediaspeech-fr-hubert")
# librosa resamples and mixes to one channel, so any file works; 16000 is
# what nearly every espnet recogniser is trained on - check this model's
# config if its audio is not 16 kHz
speech, rate = librosa.load("audio.wav", sr=16000, mono=True)
text, *_ = speech2text(speech)[0]
print(text)

RESULTS

Environments

• date: Tue Mar 22 13:50:31 UTC 2022
• python version: 3.7.11 (default, Jul 27 2021, 14:32:16) [GCC 7.5.0]
• espnet version: espnet 0.10.7a1
• pytorch version: pytorch 1.10.1
• Git hash: 1991a25855821b8b61d775681aa0cdfd6161bbc8
  • Commit date: Mon Mar 21 22:19:19 2022 +0800

asr_train_asr_hubert_raw_as_bpe150_sp

WER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
inference_asr_model_valid.acc.ave/dev_as	249	10072	49.7	41.2	9.1	7.0	57.2	100.0
inference_asr_model_valid.acc.ave/test_as	249	9920	51.1	40.1	8.9	6.5	55.4	100.0

CER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
inference_asr_model_valid.acc.ave/dev_as	249	58679	80.9	8.0	11.1	7.2	26.3	100.0
inference_asr_model_valid.acc.ave/test_as	249	58694	82.1	7.2	10.8	7.1	25.0	100.0

TER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
inference_asr_model_valid.acc.ave/dev_as	249	30837	69.5	19.0	11.5	6.3	36.8	100.0
inference_asr_model_valid.acc.ave/test_as	249	30942	70.7	17.9	11.4	6.0	35.3	100.0