CaliCastle/whisper-node

whisper-node

Node.js bindings for OpenAI's Whisper.

Features

• Output transcripts to JSON (in addition to .txt .srt .vtt)
• Runs on CPU (not GPU)
• Timestamp accurate to a single word

Installation

Your project must use typescript to continue.

1. Add dependency to project npm i whisper-node

2. Download whisper model of choice npx whisper-node download-model base.en

Usage

import whisper from 'whisper-node';

const params = {
  filePath: "example/sample.wav", // required
  model:    "medium",             // default
  output:   "JSON",               // default
}

const transcript = await whisper(params);

Sample Output

[
  {
    "tsB":    "00:00:14.310",       // time stamp begin
    "tsE":    "00:00:20.480",       // time stamp end
    "speech": "hey how's it going"  // transcription
  },
]

Made with

• Whisper OpenAI (using cpp port by: ggerganov)
• ShellJS

Roadmap

• [] Support for non-typescript projects
• [] Deprecate use of path package for browser and react-native compatibility
• [] fluent-ffmpeg to support mp3 and video ripping
• [] Pyanote diarization for speaker names