Bun-Vosk

This repository provides Vosk's capabilities directly within a Bun application.

Installation

To use this library, first ensure you have Bun installed. Then, add the necessary dependencies:

bun add bun-vosk

Usage

Setting Log Level

You can set the log level for Vosk using the setLogLevel function:

import { setLogLevel } from 'bun-vosk'

setLogLevel(0) // 0: INFO, 1: WARNING, 2: DEBUG, 3: TRACE

Model Handling

Create and manage a Vosk model using the Model class:

import { Model } from 'bun-vosk'

const model = new Model('/path/to/your/model')
model.free() // Free the model when done

Speaker Model Handling

Similarly, manage a speaker model with the SpeakerModel class:

import { SpeakerModel } from 'bun-vosk'

const speakerModel = new SpeakerModel('/path/to/your/speaker-model')
speakerModel.free() // Free the speaker model when done

Recognizer Usage

The Recognizer class provides an interface to Vosk's recognition functionalities. It can be configured for basic recognition, grammar-based recognition, or speaker recognition:

import { Model, Recognizer } from 'bun-vosk'

const model = new Model('/path/to/your/model')
const recognizer = new Recognizer({ model, sampleRate: 16000 })

// To recognize audio data
const result = recognizer.acceptWaveform(Buffer.from(audioData))

if (result) {
  console.log(recognizer.result())
}

recognizer.free()
model.free()

API

setLogLevel(level: number)

Sets the log level for Vosk.

• level: Log level (0: INFO, 1: WARNING, 2: DEBUG, 3: TRACE)

Model

Class for handling Vosk models.

• constructor(modelPath: string): Creates a new model instance.
• free(): Frees the model resources.

SpeakerModel

Class for handling Vosk speaker models.

• constructor(modelPath: string): Creates a new speaker model instance.
• free(): Frees the speaker model resources.

Recognizer

Class for handling Vosk recognition tasks.

• constructor(param: BaseRecognizerParam & T): Creates a new recognizer instance.
• setWords(words: boolean): Sets whether to include words in recognition results.
• setMaxAlternatives(max_alternatives: number): Sets the maximum number of alternative results.
• setPartialWords(partial_words: boolean): Sets whether to include partial words in recognition results.
• setSpkModel(spk_model: Model): Sets the speaker model for recognition.
• acceptWaveform(data: Buffer): Accepts audio data for recognition.
• result(): Gets the recognition result.
• partialResult(): Gets the partial recognition result.
• finalResult(): Gets the final recognition result.
• reset(): Resets the recognizer.

License

This project is licensed under the MIT License. See the LICENSE file for details.