Release v1.15

1. Implement linear_fsa_with_self_loops.
2. Allow different encoder/decoder dims for RNN-T.
3. Recover windows build.
4. Bug fixes.

Release v1.14

1. Support modified RNN-T loss.
2. Support parallel RNN-T decoding.

Release v1.13

Change rnnt_loss interface.

Release v1.12

1. Support RNN-T loss.

Release v1.11

Bugfixes for computing framewise alignments.

v1.10

Mainly bugfixes.

Release v1.9

1. Support creating ragged tensor from regular tensor.
2. Add levenshtein alignment.

Release v1.8

• Bug fixes
• Change data property of k2.RaggedTensor to values

Release v1.7

• Bug fixes
• Replace k2.RaggedInt and k2.RaggedFloat with k2.RaggedTensor
• Documentation ULR is changed to https://k2-fsa.github.io/k2/

Caution: This version is not backward compatible.

Release v1.4

Mainly bugfixes.