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
b1a1ac6c11
FunASR/egs/alimeeting/diarization/sond
History
speech_asr 2ba4683eb2 update
2023-03-16 11:14:42 +08:00
..
local add sond model 2023-02-10 18:56:14 +08:00
config_fbank.yaml add sond model 2023-02-10 18:56:14 +08:00
config.yaml add sond model 2023-02-10 18:56:14 +08:00
infer_alimeeting_test.py update 2023-03-16 11:14:42 +08:00
path.sh add sond model 2023-02-10 18:56:14 +08:00
README.md modify readme for sond 2023-02-13 11:29:35 +08:00
run.sh update 2023-03-16 11:14:42 +08:00
unit_test_modelscope.py unit test modelscope for sd 2023-02-27 16:44:06 +08:00
unit_test.py update 2023-03-16 11:14:42 +08:00

README.md

Get Started

To use this example, please execute the first stage of run.sh first to obtain the prepared data and pre-trained models:

sh run.sh --stage 0 --stop_stage 0

Then, you can execute unit_test.py to check the correctness of code:

python unit_test.py
# you will get the results:
[{'key': 'R8002_M8002_MS802-S0000_0000000_0001600', 'value': 'spk1 [(0.0, 8.88), (10.72, 11.92), (12.64, 15.2)]\nspk2 [(8.8, 9.76)]\nspk3 [(9.6, 10.96), (15.12, 15.68)]\nspk4 [(11.12, 12.72)]'}]
[{'key': 'R8002_M8002_MS802-S0000_0000000_0001600', 'value': 'spk1 [(0.0, 8.88), (10.72, 11.92), (12.64, 15.2)]\nspk2 [(8.8, 9.76)]\nspk3 [(9.6, 10.96), (15.12, 15.68)]\nspk4 [(11.12, 12.72)]'}]
[{'key': 'R8002_M8002_MS802-S0000_0000000_0001600', 'value': 'spk1 [(0.0, 8.88), (10.72, 11.92), (12.64, 15.2)]\nspk2 [(8.8, 9.76)]\nspk3 [(9.6, 10.88), (15.12, 15.68)]\nspk4 [(11.12, 12.72)]'}]
[{'key': 'test0', 'value': 'spk1 [(0.0, 8.88), (10.64, 15.2)]\nspk2 [(8.88, 9.84)]\nspk3 [(9.6, 11.04), (15.12, 15.68)]\nspk4 [(11.2, 11.76)]'}]

You can also execute run.sh to reproduce the diarization performance reported in [1]

sh run.sh --stage 1 --stop_stage 2

Results

After executing "run.sh", you will get a DER about 4.21%, which is reported in [1], Table 6, line "SOND Oracle Profile".

Reference

[1] Speaker Overlap-aware Neural Diarization for Multi-party Meeting Analysis, Zhihao Du, Shiliang Zhang, Siqi Zheng, Zhijie Yan. EMNLP 2022.