Views
No views yet
pip install chunkformer1from chunkformer import ChunkFormerModel
2
3# Load the model
4model = ChunkFormerModel.from_pretrained("khanhld/chunkFormer-rnnt-small-libri-100h")
5
6# For long-form audio transcription
7transcription = model.endless_decode(
8 audio_path="path/to/your/audio.wav",
9 chunk_size=64,
10 left_context_size=128,
11 right_context_size=128,
12 return_timestamps=True
13)
14print(transcription)
15
16# For batch processing
17audio_files = ["audio1.wav", "audio2.wav", "audio3.wav"]
18transcriptions = model.batch_decode(
19 audio_paths=audio_files,
20 chunk_size=64,
21 left_context_size=128,
22 right_context_size=128
23)1@INPROCEEDINGS{10888640,
2 author={Le, Khanh and Ho, Tuan Vu and Tran, Dung and Chau, Duc Thanh},
3 booktitle={ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
4 title={ChunkFormer: Masked Chunking Conformer For Long-Form Speech Transcription},
5 year={2025},
6 volume={},
7 number={},
8 pages={1-5},
9 keywords={Scalability;Memory management;Graphics processing units;Signal processing;Performance gain;Hardware;Resource management;Speech processing;Standards;Context modeling;chunkformer;masked batch;long-form transcription},
10 doi={10.1109/ICASSP49660.2025.10888640}}