#61 Data Preparation with prepare.sh

preview_player
Показать описание

00:00 Not all of the stages of this are necessarily relevant to all systems
01:40 Stage -1: Download LM
01:50 Stage 0: Download data
02:00 Stage 1: Prepare LibriSpeech manifest
02:30 Stage 2: Prepare musan manifest
03:00 Stage 3: Compute fbank for librispeech
05:00 Stage 4: Compute fbank for musan
05:40 Stage 5: Prepare phone based lang
08:40 Stage 6: Prepare BPE based lang
11:10 Stage 7: Prepare bigram token-level P for MMI training
14:00 Stage 8: Prepare G
14:40 Stage 9: Compile HLG
15:10 Stage 10: Compile LG
15:20 Stage 11: Generate LM training data
16:00 Stage 12: Generate LM validation data
17:18 Stage 13: Generate LM test data
17:40 Stage 14: Sort LM training data
19:49 zipformer
20:15 bpe model
23:00 bpe model stage "--byte-fallback" or "byte_fallback"
Рекомендации по теме