Vosk

Package

Speech recognition toolkit with small models that work without a network

Price: Free, open source
Access: None, runs locally

About

Kaldi-based Apache-2.0 speech recognition toolkit from Alpha Cephei. Its 50 MB models cover 20+ languages, Turkish included, and run on a Raspberry Pi, Android or servers (iOS on request), with streaming results and bindings for Python, Java, C#, Node.js and Go. Vosk Server adds WebSocket and gRPC.

What you can do with it

Transcribe microphone speech in real time with streaming partial results
Add voice commands to an Android app or a Raspberry Pi device
Generate SRT subtitles from video files with vosk-transcriber

Get started

Install the Python package with pip3 install vosk
Use 16 kHz 16-bit mono PCM WAV, or install ffmpeg for vosk-transcriber

Example

import wave
import sys

from vosk import Model, KaldiRecognizer

wf = wave.open(sys.argv[1], "rb")  # WAV format, mono PCM
model = Model(lang="en-us")

rec = KaldiRecognizer(model, wf.getframerate())
rec.SetWords(True)

while True:
    data = wf.readframes(4000)
    if len(data) == 0:
        break
    if rec.AcceptWaveform(data):
        print(rec.Result())

print(rec.FinalResult())

Details

Hosting: Self-hosted, Runs locally
Available in: Worldwide
Official SDKs: Python, Java, C#, JavaScript/TypeScript, Go
MCP server: None