Simple speech recognition using your microphone.
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.
dict Initial commit Apr 10, 2014
hmm Initial commit Apr 10, 2014
lm Initial commit Apr 10, 2014
.gitignore Initial commit Apr 10, 2014 Update readme Apr 10, 2014
apt-packages.txt Initial commit Apr 10, 2014 Initial commit Apr 10, 2014


Simple speech recognition for Python. Run the script, say some things into your microphone, and then see what you said (or an approximation).

Powered by pyaudio and Sphinx.



Download sphinxbase and extract the files.

Now, run:

cd sphinxbase
./configure;make clean all;make install
cd python
python install

You may need to use sudo for make install or python install.


Download pocketsphinx and extract the files.

Now, run:

cd pocketsphinx
./configure;make clean all;make install
cd python
python install

Packages (Linux only)

Now, run:

cd speech-recognizer
sudo xargs -a apt-packages.txt apt-get install


Now, download the right version of pyaudio and install it.

Language files

If you want to speak english, you need to get the english language model and the english acoustic model.

You will need to put the acoustic model into scribe/hmm, and the language model into scribe/lm.

The filetree should look like this for english:

├── dict
│   └── cmu07a.dic
├── hmm
│   ├── feat.params
│   ├── feature_transform
│   ├── mdef
│   ├── means
│   ├── mixture_weights
│   ├── noisedict
│   ├── README
│   ├── transition_matrices
│   └── variances
├── lm
│   └── cmusphinx-5.0-en-us.lm.dmp

For other languages, check here, or see below on training your own model. If you use different language models, acoustic models, or dictionaries, you will want to change these paths in

HMDIR = os.path.join(BASE_PATH, "hmm")
LMDIR = os.path.join(BASE_PATH, "lm/cmusphinx-5.0-en-us.lm.dmp")
DICTD = os.path.join(BASE_PATH, "dict/cmu07a.dic")


To run, you just have to:

cd speech-recognizer

You should be able to talk for a few seconds, after which it will spend some time processing, and the show you what you said.


There are some options that you can modify at the top of The easiest one to modify is RECORD_SECONDS.

More reading

To find out more, read up on sphinx.

You can train the language models to make them more accurate, use unsupported languages, or be more domain-specific.