xbmu_amdo31

This is the official repository of the XBMU-AMDO31 dataset. Speech database of Tibetan Amdo dialect for speech recognition.

XBMU-AMDO31 version: 1.0.0 (12/03/2022)

Download

1. Method 1: Please download from openslr
2. Method 2: Please download from huggingface

Leaderboard

Contributor	Toolkit	Train Recipe	Features	Modeling unit	Inference	Dev/Test WER
Baseline	Kaldi	NNET3	Fbank	Syllables	model example	17.71 / 15.29
Baseline	Espnet	[Conformer/Transformer-AED]	Fbank	Syllables	model example	14.80 / 13.80
Baseline	Espnet	[Conformer/Transformer-AED]	Fbank	Alphabet	model example	18.00 / 16.30
Baseline	Espnet	[Conformer/Transformer-AED]	Fbank	BPE500	model example	10.60 / 10.10
Baseline	Espnet	[Conformer/Transformer-AED]	HUBERT	BPE500	model example	9.80 / 9.10
Baseline	Espnet	[Conformer/Transformer-AED + wenetspeech pre-trained model]	HUBERT	BPE500	model example	9.90 / 9.60
Baseline	Espnet	[Conformer/Transformer-AED + wenetspeech pre-trained model]	Fbank	BPE500	model example	9.30 / 8.90
Baseline	Icefall	Pruned Stateless5	Fbank	BPE500	model	11.24 / 10.57
Baseline	Icefall	Pruned Stateless7	Fbank	BPE500	model	10.12 / 9.70