vosk

What it is and what it does

Vosk is a Python binding for an offline speech recognition engine based on Kaldi. It transcribes audio to text without sending data to external servers, making it suitable for privacy-sensitive applications, embedded systems, and environments without reliable internet. The package wraps compiled Kaldi models and provides a streaming API that processes audio chunks in real time, returning partial and final transcriptions as they become available.

The toolkit supports 20+ languages and dialects with relatively small model files (around 50 Mb each). It's designed to scale from resource-constrained devices like Raspberry Pi to larger deployments, and includes features like reconfigurable vocabulary and speaker identification. Common use cases include chatbot backends, smart home voice control, video subtitle generation, and lecture transcription.

Use it for:

• Build a voice-controlled chatbot or virtual assistant that processes audio locally without cloud dependency
• Generate subtitles for video or audio recordings by transcribing speech to text offline
• Add voice commands to a Raspberry Pi or embedded Linux device for IoT applications
• Transcribe interviews or lectures with privacy guarantees and no per-minute API costs
• Implement speaker identification or custom vocabulary matching in voice-driven workflows

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Vosk provides offline speech recognition for 20+ languages, enabling real-time transcription via a streaming API without requiring cloud services or internet connectivity.

Yes, if you need offline speech recognition and can tolerate medium install friction and model file management. The permissive license, active maintenance, and lack of known vulnerabilities make it a solid choice for privacy-focused or bandwidth-constrained deployments. Not recommended if you require state-of-the-art accuracy or support for languages beyond the 20+ listed.

Install

vosk on PyPI

pip install vosk

uv add vosk

poetry add vosk

Installing vosk

Before you install

Medium install friction due to platform-specific wheels and compiled dependencies via cffi. Active maintenance with recent commits and strong community engagement (15056 GitHub stars), though the latest release was 2022-12-14.

License in practice

Licensed under Apache Software License (permissive), allowing commercial and private use with minimal restrictions—suitable for most production and proprietary projects.

Quickstart

pip install vosk

import vosk
from vosk import Model, KaldiRecognizer

model = Model(lang="en-us")
recognizer = KaldiRecognizer(model, samplerate)

Requires a Vosk model file (approximately 50 Mb per language) to be downloaded separately and available on disk; audio input source must be configured.

Verify before relying

• Whether models for all 20+ languages are actively maintained and available for download
• Performance characteristics (latency, accuracy) compared to cloud-based alternatives
• Exact sample rate and audio format requirements for the streaming API

Package facts

Evidence: vosk-0.3.45-py3-none-linux_armv7l.whl; vosk-0.3.45-py3-none-manylinux2014_aarch64.whl; vosk-0.3.45-py3-none-manylinux_2_12_x86_64.manylinux2010_x86_64.whl; vosk-0.3.45-py3-none-win_amd64.whl

Tags

More Python Modules packages