On January 6, the release of 1.7.4 of the high-performance automatic speech recognition system took place whisper.cpp, implementing the model Whisper from OpenAI, written in C and C++ and licensed under the MIT license.
The project uses the author's tensor machine learning library GGML, written in C and optimized for using various backends (CUDA, Vulkan, BLAS, SYCL, OpenCL, etc.).
List of changes:
- Renaming executable files:
main -> whisper-cli bench -> whisper-bench stream -> whisper-stream command -> whisper-command server -> whisper-server talk-llama -> whisper-talk-llama
- Removed talk and talk.wasm examples.
- An option to suppress non-speech lexemes has been added to whisper-server.
- The suppress_non_speech_tokens parameter has been renamed to suppress_nst.
- The no_speech_thold option has been added to the API and utilities to set the probability of no speech at the segment level during transcription.
- Added --suppress_nst option to whisper-cli.
- Other improvements and bug fixes.
Source: linux.org.ru
