llamaindex-whisper
v0.0.5
Published
Whisper reader for llamaindex
Downloads
7
Readme
LlamaIndex integration with OpenAI's whisper
Installation
This integrates the LlamaIndexTS project with OpenAI's whisper speech transcription and translation library.
Required libraries
To start, ensure you have whisper and ffmpeg locally:
pip install openai-whisper
brew install ffmpeg
And this package, of course:
npm install llamaindex-whisper
Getting started
Simply import the WhisperReader
and load your data. It's that easy!
loadData
returns an array of LlamaIndex Document objects. The array always contains one Document with all of the text. (Support for chunking and splitting is a future concern.)
import { WhisperReader } from "llamaindex-whisper";
const reader = new WhisperReader();
const documents = reader.loadData("this-is-water-speech.mp3");
// => Document(text="Greetings parents and congratulations to Kenyon’s graduating class of 2005…")
Combining that with llamaindex itself:
import { VectorStoreIndex } from "llamaindex";
import { WhisperReader, WhisperDevice, WhisperOutputFormat } from "llamaindex-whisper";
async function main() {
const whisperReader = new WhisperReader({
device: WhisperDevice.CPU,
outputFormat: WhisperOutputFormat.Text,
});
const documents = await whisperReader.loadData("./giant-leap.mp3")
// Split text and create embeddings. Store them in a VectorStoreIndex
const index = await VectorStoreIndex.fromDocuments(documents);
// Query the index
const queryEngine = index.asQueryEngine();
const response = await queryEngine.query(
"How big were the step and the leap?"
);
// Output response
console.log(response.toString());
// => Based on the given context information, the step and the leap are not described in terms of their size or magnitude.
// Well, you can't win them all…
}
main();
Options
WhisperReader
supports a subset of the whisper CLI. Here are the ones supported so far with their defaults:
model: WhisperModel = WhisperModel.Base;
temperature: number = 0;
language: WhisperLanguage = WhisperLanguage.English;
outputDirectory = ".";
outputFormat = WhisperOutputFormat.All;
task: WhisperTask = WhisperTask.Transcribe;
device: WhisperDevice = WhisperDevice.CUDA;
Model
WhisperModel
could be any of the supported whisper models and defaults to Base
. English
variants are specifically fine-tuned for english processing. The generic models are multilingual:
TinyEnglish
Tiny
BaseEnglish
Base
SmallEnglish
Small
MediumEnglish
Medium
LargeV1
LargeV2
Large
Language
WhisperLanguage
has mappings for nearly 100 languages, from Afrikaans to Yoruba. Specifically:
Afrikaans, Albanian, Amharic, Arabic, Armenian, Assamese, Azerbaijani, Bashkir, Basque, Belarusian, Bengali, Bosnian, Breton, Bulgarian, Catalan, Chinese, Croatian, Czech, Danish, Dutch, English, Estonian, Faroese, Finnish, French, Galician, Georgian, German, Greek, Gujarati, HaitianCreole, Hausa, Hawaiian, Hebrew, Hindi, Hungarian, Icelandic, Indonesian, Italian, Japanese, Javanese, Kannada, Kazakh, Khmer, Korean, Lao, Latin, Latvian, Lingala, Lithuanian, Luxembourgish, Macedonian, Malagasy, Malay, Malayalam, Maltese, Maori, Marathi, Mongolian, Myanmar, Nepali, Norwegian, Nynorsk, Occitan, Pashto, Persian, Polish, Portuguese, Punjabi, Romanian, Russian, Sanskrit, Serbian, Shona, Sindhi, Sinhala, Slovak, Slovenian, Somali, Spanish, Sundanese, Swahili, Swedish, Tagalog, Tajik, Tamil, Tatar, Telugu, Thai, Tibetan, Turkish, Turkmen, Ukrainian, Urdu, Uzbek, Vietnamese, Welsh, Yiddish, Yoruba
OutputFormat
WhisperOutputFormat
tells whisper the format of the file it writes for transcription. The default is All
, which writes one file for each of the supported types. I don't know why they decided on that—I can't imagine it's what people want—so you'll want to specify an output.
All
Text
VTT
SRT
TSV
JSON
Task
WhisperTask
can be set to either Transcribe
or Translate
. Transcription is the default.
Device
WhisperDevice
can be either CPU
or CUDA
. This defaults to CUDA
, so take note to swap into CPU mode if your graphics card does not support it.