Master Python Speech Recognition: Install, Record, and Transcribe Audio

Speech recognition originated in the early 1950s at Bell Labs; early systems could only recognize a single speaker and a limited vocabulary, while modern systems handle multiple speakers and many languages.

How Speech Recognition Works

Audio captured by a microphone is converted to an electrical signal, digitized, and then processed by models such as Hidden Markov Models (HMM) or neural networks to produce text.

Choosing a Python Speech‑Recognition Package

PyPI provides several packages, including apiai, google-cloud-speech, pocketsphinx, SpeechRecognition, watson-developer-cloud, and wit. The SpeechRecognition library is highlighted for its ease of use.

Installing SpeechRecognition

$ pip install SpeechRecognition

Verify the installation:

>> import speech_recognition as sr
>>> sr.__version__
'3.8.1'

Using the Recognizer Class

Create a Recognizer instance and call its various recognize_* methods (Google, Bing, IBM, etc.). The available methods are:

recognize_bing()

recognize_google()

recognize_google_cloud()

recognize_houndify()

recognize_ibm()

recognize_sphinx()

recognize_wit()

Working with Audio Files

Download an audio file (e.g., harvard.wav) and place it in the working directory. Use AudioFile with a context manager to obtain an AudioData instance.

harvard = sr.AudioFile('harvard.wav')
with harvard as source:
    audio = r.record(source)
print(type(audio))

Transcribe the whole file:

print(r.recognize_google(audio))

Limit recording duration or offset to capture specific segments:

with harvard as source:
    audio = r.record(source, duration=4)
    print(r.recognize_google(audio))

with harvard as source:
    audio = r.record(source, offset=4, duration=3)
    print(r.recognize_google(audio))

Handling Ambient Noise

Before recording, call adjust_for_ambient_noise to calibrate the noise floor. The default analyzes one second; you can shorten it with the duration argument.

with harvard as source:
    r.adjust_for_ambient_noise(source, duration=0.5)
    audio = r.record(source)
    print(r.recognize_google(audio))

Microphone Input

Install PyAudio (required for microphone access). Installation varies by OS:

Debian/Ubuntu: $ sudo apt-get install python-pyaudio python3-pyaudio then $ pip install pyaudio macOS: $ brew install portaudio then $ pip install pyaudio Windows: $ pip install pyaudio Capture live speech:

with sr.Microphone() as source:
    r.adjust_for_ambient_noise(source)
    audio = r.listen(source)
    print(r.recognize_google(audio))

Dealing with Unrecognizable Speech

Wrap recognition calls in try/except blocks to catch speech_recognition.UnknownValueError when the audio cannot be transcribed.

Transcribing Other Languages

All recognize_* methods accept a language keyword; set it to the appropriate language code to transcribe non‑English speech.