CMU Sphinx
cmusphinx.github.io
- Free plan
- Windows
- macOS
- Linux
- Android
- Self-hosted
What CMU Sphinx is
CMU Sphinx is an open-source toolkit for developers building speech applications. Its maintained components are PocketSphinx, a lightweight C speech-recognition library, and SphinxTrain, which provides acoustic-model training tools. Designed for efficient recognition on low-resource platforms, it supports keyword spotting, grammar and language-model searches, and word-level segmentation. Recognition needs acoustic and language models; the project lists prebuilt models for languages including English, Chinese, French, Spanish, German, and Russian. SphinxTrain offers a process for adding a language by preparing data, training a model, and testing it. Noise reduction uses MFCC features, noise tracking, and spectral subtraction. The toolkit supports offline, on-device recognition, and its license is described as BSD-like with commercial distribution allowed. Documented application ideas include voice control, language learning, transcription, closed captioning, speech translation, and voice search. Decoders do not convert encoded audio formats, so input audio must first be converted to PCM.
Who it suits
It suits developers applying speech recognition in applications, rather than researchers focused on speech recognition itself. It is relevant where efficient or offline recognition is needed on low-resource platforms.
Good to know, and what to watch
- Supports keyword spotting and grammar searches.
- Provides word-level segmentation.
- Designed for low-resource platforms.
- Supports offline, on-device recognition.
- BSD-like license permits commercial distribution.
- Recognition requires acoustic and language models.
- Audio must be converted to PCM first.
- Large-vocabulary recognition is too demanding for small devices.
- The Android tutorial may not work with current tools.
Plans
Every plan and priceCMU Sphinx toolkit
Open source
Getting started
- Visit the CMU Sphinx website.
- Choose PocketSphinx for recognition or SphinxTrain for acoustic model training.
- Prepare acoustic and language models for the language you need.
- Convert input audio to PCM before processing.
- For Android, consult the documented PocketSphinx library and demo, bearing in mind the tutorial may not work with current tools.
Limits to check
Recognition needs acoustic and language models, and input audio must be converted to PCM. On phones and other small embedded devices, large-vocabulary recognition is too computationally demanding; limited vocabularies are commonly used.
The verdict
CMU Sphinx provides developers with recognition and acoustic-model training components for speech applications. Plan for model requirements and PCM audio preparation, and consider the documented limits for small devices and Android tooling.
Questions people ask
Does CMU Sphinx cost anything?
The toolkit is listed as free. Its BSD-like license allows commercial distribution.
Which components are maintained?
PocketSphinx is a lightweight C recognizer library, and SphinxTrain provides acoustic model training tools.
Which languages can it recognize?
Recognition requires acoustic and language models. Prebuilt models are listed for English, Chinese, French, Spanish, German, and Russian.
Can it run offline or on a device?
The toolkit documents offline and on-device recognition, including a PocketSphinx Android library and demo.
Can it process encoded audio directly?
No. Audio must be converted to PCM before the decoders process it.
What does the Android documentation say?
The site documents a PocketSphinx Android library and demo, while cautioning that the tutorial has not been tested for quite some time and may not work with current Android tools.
What the maker publishes
- Android support
- The site documents a PocketSphinx Android library and demo, while cautioning that the tutorial has not been tested for quite some time and may no longer work with current Android tools. source
- Commercial use
- The site describes its license as BSD-like and says it allows commercial distribution. source
- Current release
- The homepage announces PocketSphinx 5.1.0, released May 6, 2026, with Python 3.14 support and prebuilt Linux/arm64 wheels. source
- Developer audience
- The developer tutorial is intended for developers applying speech technology in applications, rather than speech-recognition researchers. source
- Input audio requirement
- The decoders do not convert encoded audio formats, so audio must be converted to PCM before processing. source
- Languages
- CMU Sphinx is language-independent, but recognition requires acoustic and language models; the site lists prebuilt models for languages including English, Chinese, French, Spanish, German, and Russian. source
- Low-resource design
- The tools are designed for efficient speech recognition on low-resource platforms. source
- Maintained components
- The currently maintained components are PocketSphinx, a lightweight C recognizer library, and SphinxTrain, acoustic model training tools. source
- Model training
- SphinxTrain provides acoustic model training tools, and the toolkit describes collecting data, cleaning it, training a model, and testing it to add a language. source
- Noise handling
- CMU Sphinx uses mel-cepstrum MFCC features with noise tracking and spectral subtraction for noise reduction. source
- Platform constraints
- The FAQ says large-vocabulary speech recognition is too computationally demanding for phones and other small embedded devices, where limited vocabularies are commonly used. source
- Purpose
- CMU Sphinx is an open-source speech recognition toolkit for building speech applications. source
- Recognition features
- The toolkit supports keyword spotting, grammar searches, language-model searches, and word-level segmentation through its documented interfaces. source
- Support
- The site describes commercial support and directs users to GitHub project and issue trackers for help. source
- Use cases
- The tutorial lists voice control, language learning, transcription, closed captioning, speech translation, and voice search as possible applications. source
Checked against the maker's pages on 5 Oct 2026.