LlamaEdge-Whisper 0.3.1

Major changes:

• New CLI options:

  • threads: Number of threads to use during computation. Defaults to 4.
  • processors: Number of processors to use during computation. Defaults to 1.
  • task: Task type. Default to full. Possible values: transcribe, translate, full
  • port: Port number. Defaults to 8080.

• Support new fields of transcription requests

  • language: The language of the input audio. Defaults to en.
  • temperature: Sampling temperature, between 0 and 1. Defaults to 0.00.
  • prompt: Text to guide the model's style or continue a previous audio segment. Defaults to none
  • max_len: Maximum number of tokens that the model can generate in a single transcription segment (or chunk). Defaults to 0.
  • split_on_word: Split audio chunks on word rather than on token. Defaults to false.
  • detect_language: Automatically detect the spoken language in the provided audio input. Defaults to false.
  • offset_time: Time offset in milliseconds. Defaults to 0.
  • duration: Length of audio (in seconds) to be processed starting from the point defined by the offset_time field (or from the beginning by default). Defaults to 0.

• Support new fields for translation requests

  • detect_language: Automatically detect the spoken language in the provided audio input. Defaults to false.
  • offset_time: Time offset in milliseconds. Defaults to 0.
  • duration: Length of audio (in seconds) to be processed starting from the point defined by the offset_time field (or from the beginning by default). Defaults to 0.
  • max_len: Maximum number of tokens that the model can generate in a single transcription segment (or chunk). Defaults to 0.
  • split_on_word: Split audio chunks on word rather than on token. Defaults to false.