LlamaEdge-Whisper 0.3.9

Major changes:

• (New) Support API key (see the start server section of README for details)
• (New) Provide the /v1/files endpoint for uploading, removing, and listing files
• Upgrade dependencies:
  • wavup v0.1.5
  • endpoints v0.23.2
  • llama-core v0.25.3

LlamaEdge-Whisper 0.3.8

Major changes:

• Upgrade wavup to v0.1.4
  • Improve the support for multi-channel audio in trim_ending_silence func
• Upgrade llama-core to v0.24.1
  • Keep the metadata state of whisper-api-server same as whisper plugin

LlamaEdge-Whisper 0.3.7

Major changes:

• Added the use_new_context field in the requests of transcriptions and translations to decide if creating a new whisper computation context for each request

LlamaEdge-Whisper 0.3.6

Major changes:

• Improve language and detect_language request params
• Upgrade wavup dep to 0.1.3
  • Improve the design of trimming tailing silence audio samples

LlamaEdge-Whisper 0.3.5

Major changes:

• Upgrade wavup dep to optimize the trimming process

LlamaEdge-Whisper 0.3.3

Major changes:

• Support input audio files

  • formats: caf, isomp4, mkv, ogg, aiff, wav
  • codecs: aac, adpcm, alac, flac, mp1, mp2, mp3, pcm, vorbis
  • Fixed #3

• Trim the ending silence

  • Fixed #4

LlamaEdge-Whisper 0.3.2

Major changes:

• New endpoints

  • GET /v1/files/{file_id}: Retrieve information of a specific file by id
  • DELETE /v1/files/{file_id}: Remove a specific file by id

• Upgrade to llama-core v0.22.0

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.

LlamaEdge-Whisper 0.3.0

Major change:

• Remove the support for /v1/audio/speech endpoint. The endpoint will be supported in the coming tts-api-server.

LlamaEdge-Whisper 0.2.2

Major change:

• Migrate to WasmEdge v0.14.1