EngineX-Iluvatar/enginex_bi_series-sherpa-onnx
Archived
16
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

Add fine-tuned whisper model on aishell (#565)

Browse Source 
See also https://github.com/k2-fsa/icefall/pull/1466
This commit is contained in:
Fangjun Kuang
2024-01-31 17:23:42 +08:00
committed by GitHub
parent 0b18ccfbb2
commit 2e8b321210
3 changed files with 36 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
scripts/whisper/test.py
View File

@@ -257,9 +257,9 @@ def compute_features(filename: str) -> torch.Tensor:
mel = (log_spec + 4.0) / 4.0
# mel (T, 80)
# We pad 50 frames at the end so that it is able to detect eot
# You can use another value instead of 50.
mel = torch.nn.functional.pad(mel, (0, 0, 0, 1000), "constant", 0)
# We pad 1500 frames at the end so that it is able to detect eot
# You can use another value instead of 1500.
mel = torch.nn.functional.pad(mel, (0, 0, 0, 1500), "constant", 0)
# Note that if it throws for a multilingual model,
# please use a larger value, say 300