Open In App

Speech Recognition in Python using CMU Sphinx

Last Updated : 01 Dec, 2022
Like Article

“Hey, Siri!”, “Okay, Google!” and “Alexa playing some music” are some of the words that have become an integral part of our life as giving voice commands to our virtual assistants make our life a lot easier. But have you ever wondered how these devices are giving commands via voice/speech?

Do applications understand your voice? How does the computer even decode this if it only understands 0/1?

The answer is simple: it uses Speech Recognition software to decode the user input received as speech/voice using the device’s microphone. Speech Recognition software to decode the user input received as speech/voice using the device’s microphone. the task of this software is to convert the speech to a string(text) so that the computer can then decode it.

One such Toolkit is CMU Sphinx which is an open-source toolkit used for speech recognition, it also has a lightweight recognizer library called Pocketsphinx which will be used to recognize the speech. This library is a great resource especially when you are offline as when you have internet access you should prefer Google API with speech recognition due to higher precision. but when you are building a project that works offline or uses speech on an offline embedded device, use pocketsphinx.

Recognition Process

Let’s discuss how this library works from behind to actually recognize our voice, It takes a waveform and then splits it according to utterances by silence then traverses and tries to find out what is being said in each utterance for accomplishing this task it takes all possible combinations of words and try to match them with audio choosing the best matching combination.

Installation of modules

Since pocketsphinx is an external library i.e. its not present as an inbuilt entity in python we would install it to our machines using pip installer and then using import to invoke all the functionalities of this library,

Now open your terminal and type the following command 

NOTE- make sure that you have latest version of pip installed if not then type following

python -m pip install --upgrade pip setuptools wheel

If you have latest version of pip then proceed directly and type the following code into your terminal.

pip install pocketsphinx

Now that you have installed pocketsphinx in your machine lets move forward to more.


There are two prerequisite library which is used along side with pocketsphinx they are :-

  1. SpeechRecognition – used for speech recognition ,with support for several engines and APIs, online and offline.
  2. PyAudio-used to play and even record audio in python.

Now it is recommended to install these two library using pip install command:-

pip install SpeechRecognition
brew install portaudio
pip install pyaudio

Now installation of all required external library is completed so lets move forward to code.


It is an external iterator class available in pocketsphinx which can be used for continuous recognition or keyword search from a microphone.
Here is the code for continuous recognition.


# import LiveSpeech
from pocketsphinx import LiveSpeech
for phrase in LiveSpeech():
    # here the result is stored in phrase which
    # ultimately displays all the words recognized
    print("Sorry! could not recognize what you said")

Output :   



We used LiveSpeech in a basic for in loop to fetch continuous speech input from user using the device microphone then we store the converted string into phrase and display each word uttered by the user.

Keyword searching

We use an variable named speech of type pocketsphinx.LiveSpeech , In which we invoke the class LiveSpeech with arguments keyphrase i.e. the keyword to be searched and kws_threshold then we used an for in loop on speech which continuously looks for user input in form of voice if the user utters the word ‘forward’  then it is printed along with segments.


# importing livespeech
from pocketsphinx import LiveSpeech
speech = LiveSpeech(keyphrase='forward', kws_threshold=1e-20)
# an for in loop to iterate in speech
for phrase in speech:
        # printing if the keyword is spoken with segments along side.

Output :



Test program

First of all import speech_recognition with referencing it as some reference name aud now you can recognize speech using your code.

Now fetch audio from devices microphone and store in variable reference of type speech_recognition.Recognizer to recognize the audio and convert to text. After that define microphone as your source of input and define an variable reference say audio to listen i.e it takes user input of speech and stores it there, then we use invoke sphinx using try we try printing what user said here we invoke recognize_sphinx and pass argument audio, now the work of this class to convert what user said (in form of speech ) to text form and display it in console simply called Recognition.

If the code is unable to accept voice input due to unclear voice then we throw an exception for unclear voice and for RequestError tool.


import speech_recognition as aud
# fetch audio from devices microphone
# and store in variable reference of type speech_recognition
a = aud.Recognizer()
# declaring device microphone as the source to take audio input
with aud.Microphone() as source:
    print("Say something!")
    # variable audio prints what user said in text format the end
    audio = a.listen(source)
# invoking sphinx for speech recognition
    # printing audio
    print("You said " + a.recognize_sphinx(audio))
except aud.UnknownValueError:
    # if the voice is unclear
    print("Could not understand")
except aud.RequestError as e:
    print("Error; {0}".format(e))





This winds up our topic of discussion of Speech recognition using CMU Sphinx , there lot of more applications of this useful library.

Like Article
Suggest improvement
Share your thoughts in the comments

Similar Reads