speech_recognition/FunASR
1
0
Fork 0
mirror of https://github.com/modelscope/FunASR synced 2025-09-15 14:48:36 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

TOLD/SOND: update finetune.sh

Browse Source
This commit is contained in:
志浩 2023-08-01 20:53:55 +08:00
parent 4fdcba65bb
commit 16c834bcf8
2 changed files with 12 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
egs/callhome/diarization/sond/finetune.sh
View File

@ -221,13 +221,17 @@ if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
done
fi
# You will get a DER like:
# iter0: 9.68 10.51
# iter1:
# iter2:
# iter3:
# iter4:
# In this stage, we need the raw waveform files of Callhome corpus.
# Due to the data license, we can't provide them, please get them additionally.
# And convert the sph files to wav files (use scripts/dump_pipe_wav.py).
# Then find the wav files to construct wav.scp and put it at data/callhome2/wav.scp.
# After iteratively perform SOAP, you will get DER results like:
# iters| oracle_vad | system_vad
# iter_0: 9.68 | 10.51
# iter_1: 9.26 | 10.14 (reported in the paper)
# iter_2: 9.18 | 10.08
# iter_3: 9.24 | 10.15
# iter_4: 9.27 | 10.17
if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
for dset in ${test_sets}; do
echo "stage 4: Evaluating finetuned system on ${dset} set with medfilter_size=83 clustering=EEND-OLA"

2
egs/callhome/diarization/sond/script/extract_nonoverlap_segments.py
View File

@ -93,7 +93,7 @@ def process(task_args):
extracted_spk = []
count = 1
for st, ed, spk in tqdm(turns, total=len(turns), ascii=True, disable=args.no_pbar):
if (ed - st) >= args.min_dur * args.sr and wav[st: ed] >= args.min_dur * args.sr:
if (ed - st) >= args.min_dur * args.sr and len(wav[st: ed]) >= args.min_dur * args.sr:
seg = wav[st: ed]
save_path = os.path.join(args.out_dir, mid, spk, "{}_U{:04d}.wav".format(spk, count))
if not os.path.exists(os.path.dirname(save_path)):