Releases: k2-fsa/sherpa
Releases · k2-fsa/sherpa
doc
Release v1.2
- Support pre-trained CTC models from NeMo
Release v1.0
Release v1.0 (#186) * Release v1.0 * Fix building on windows
Release v0.9.1
Support LSTM transducer.
Release v0.8
Support endpointing.
Release v0.7
Add c++ frontend for non-streaming ASR.
See https://k2-fsa.github.io/sherpa/cpp/installation/index.html for installation.
See https://k2-fsa.github.io/sherpa/cpp/offline_asr/index.html for usage.
Note: It supports conda install
on Linux/macOS/Windows.
Release v0.6
Bugfixes.
Release v0.5
- Add
@torch.no_grad()
to other threads in addition to the main thread to reduce memory consumption sincetorch.no_grad()
is thread-local
Release v0.4
- Support torch >= 1.6.0
- Support specifying the max number of clients that can be handled by the server
Release v0.3
- Support streaming ASR with pruned stateless Emformer RNN-T
- Add checks for torch version during import time
- Support Windows