diff --git a/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/.train_adapter.py.swp b/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/.train_adapter.py.swp index 2d201c0c3..5f6b4aa62 100644 Binary files a/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/.train_adapter.py.swp and b/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/.train_adapter.py.swp differ diff --git a/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/train_adapter.py b/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/train_adapter.py index 32fad2147..4901d5fb7 100755 --- a/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/train_adapter.py +++ b/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/train_adapter.py @@ -834,6 +834,7 @@ def compute_loss( texts = batch["supervisions"]["text"] print(texts) + #texts = batch["greedy pseudo text"] token_ids = sp.encode(texts, out_type=int) y = k2.RaggedTensor(token_ids).to(device) @@ -1582,6 +1583,14 @@ def run_adapter(rank, world_size, args, wb=None): #train_cuts = librispeech.train_clean_10_cuts(option='male') train_cuts = librispeech.test_clean_user(option='big') + pl_texts = open('/workspace/icefall/egs/librispeech/ASR/tc_big.pl', 'r').readlines() + pl = {} + for text in pl_texts: + text = text.strip().split() + pl[text[0]] = ' '.join(text[1:]) + + print(pl) + exit() def remove_short_and_long_utt(c: Cut): return 1.0 <= c.duration <= 20.0