Skip to content

Releases: fquirin/speech-recognition-experiments

setup files + models + v1.0.0 code for whisper-tflite, whisper-cpp, sherpa-ncnn

19 Jan 19:49
@fquirin fquirin
v1.0.0
00048ef
Compare
Choose a tag to compare
setup files + models + v1.0.0 code for whisper-tflite, whisper-cpp, sherpa-ncnn Latest
Latest

Added 3 engines

  • whisper-tflite (+ tiny tflite model)
  • whisper-cpp (+ tiny ggml model)
  • sherpa-ncnn (+ small model)

[EDIT 2023.02.16]

  • Added converted tiny model for CTranslate2 (whisper-tiny-ct2.zip)
  • Added 'tflite_runtime' wheel for arm64 (Python 3.9) built with Bazel (much faster runtime!)

[EDIT 2023.03.04]

  • Added KenLM binaries for amd64 and aarch64
  • Added OpenSeq2Seq 'ctc-decoders' wheels for amd64 and aarch64 (Python 3.9)

[EDIT 2023.03.12]

  • Added KenLM binareis for armv7l (just in case) and wheels for Python bindings
  • Added Flashlight 'text' + 'sequence' wheels for amd64 and aarch64

[EDIT 2023.05.09]

  • Updated whisper-tiny-ct2 model (int8)
  • Added whisper-small-ct2 model (int8)

[EDIT 2023.05.19]

  • Added tokenizer to whisper-tiny-ct2 and small model, so they don't need to be downloaded at first run
Assets 20