Python Speech Recognition

In this article, we will learn how to perform Speech Recognition in Python.

Speech Recognition and Speech-to-Text are often used interchangeably. But Speech-to-Text is only a subfield of Speech Recognition. Other forms of Speech Recognition include Wake Word Detection, Voice Command Recognition, and Voice Activity Detection (VAD).

Below is the cheat sheet I use when deciding which Speech Recognition algorithm to use:

• Do you need to detect if a person is talking and when? Then use Cobra Voice Activity Detection.
• Do you need to detect the occurrence of a single phrase? Or one of a few phrases? Then Porcupine Wake Word is the right engine.
• Do you need to understand voice commands? Rhino Speech-to-Intent is the correct tool here. Rhino can infer users' intent and accurately extract the request's details (i.e., slot values) using minimum runtime resources.
• Do you need to transcribe speech to text in real time? Use Cheetah Streaming Speech-to-Text.
• Do you need to transcribe large volumes of speech to text in batch mode? Leopard Speech-to-Text is the right tool.

Cobra Voice Activity Detection

1- Install the Cobra Voice Activity Detection SDK using PIP:

pip3 install pvcobra

2- Sign up for a free Picovoice Console account and copy your AccessKey. It handles authentication and authorization.

3- Create an instance of the Voice Activity Detection engine:

import pvcobra

cobra = pvcobra.create(access_key='${ACCESS_KEY}')

4- Pass in frames of audio to the .process method:

while True:
    audio_frame = get_next_audio_frame()
    voice_probability = cobra.process(audio_frame)

For more information check Cobra Voice Activity Detection's product page or refer to Cobra's Python SDK quick start guide.

Porcupine Wake Word

1- Install the Porcupine Wake Word SDK using PIP:

pip3 install pvporcupine

2- Sign up for a free Picovoice Console account and copy your AccessKey. It handles authentication and authorization.

3- Create your custom wake word model using Picovoice Console.

4- Create an instance of the Wake Word engine:

porcupine = pvporcupine.create(
  access_key='${ACCESS_KEY}',
  keyword_paths=['${KEYWORD_FILE_PATH}']
)

5- Pass in frames of audio to the .process method:

while True:
  audio_frame = get_next_audio_frame()
  keyword_index = porcupine.process(audio_frame)
  if keyword_index > -1:
      # keyword detected
      pass

For more information check Porcupine Wake Words's product page or refer to Porcupine's Python SDK quick start guide.

Rhino Speech-to-Intent

1- Install the Rhino Speech-to-Intent SDK using PIP:

pip3 install pvrhino

2- Sign up for a free Picovoice Console account and copy your AccessKey. It handles authentication and authorization.

3- Create your Context using Picovoice Console.

4- Create an instance of Rhino Speech-to-Intent to start recognizing voice commands within the domain of the provided context:

import pvrhino

rhino = pvrhino.create(
   access_key='${ACCESS_KEY}',
   context_path='${CONTEXT_FILE_PATH}'
)

5- Pass in frames of audio to the .process function and use the .get_inference function to determine the user's intent:

while True:
   audio_frame = get_next_audio_frame()
   is_finalized = rhino.process(audio_frame)
   if is_finalized:
      # get inference if is_finalized is true
      inference = rhino.get_inference()
      if inference.is_understood:
         # use intent and slots if inference was understood
         intent = inference.intent
         slots = inference.slots

For more information check Rhino Speech-to-Intent's product page or refer to Rhino's Python SDK quick start guide.

Cheetah Streaming Speech-to-Text

1- Install the Cheetah Streaming Speech-to-Text SDK using PIP:

pip3 install pvcheetah

2- Sign up for a free Picovoice Console account and copy your AccessKey. It handles authentication and authorization.

3- Create an instance of Cheetah to transcribe speech to text in real-time:

import pvcheetah

cheetah = pvcheetah.create(access_key='${ACCESS_KEY}')

4- Pass in audio frames as they become available to the .process function:

while True:
    partial_transcript, is_endpoint = cheetah.process(get_next_audio_frame())
    if is_endpoint:
        final_transcript = cheetah.flush()

For more information check Cheetah Streaming Speech-to-Text's product page or refer to Cheetah's Python SDK quick start guide.

Leopard Speech-to-Text

1- Install the Leopard Speech-to-Text SDK using PIP:

pip3 install pvleopard

2- Sign up for a free Picovoice Console account and copy your AccessKey. It handles authentication and authorization.

3- Create an instance of Leopard to transcribe speech to text:

import pvleopard

leopard = pvleopard.create(access_key='${ACCESS_KEY}')

4- Pass in an audio file to Leopard and inspect the result:

transcript, words = leopard.process_file('${AUDIO_PATH}')
print(transcript)
for word in words:
    print(
      "{word=\"%s\" start_sec=%.2f end_sec=%.2f confidence=%.2f}"
      % (word.word, word.start_sec, word.end_sec, word.confidence))

For more information, check Leopard Speech-to-Text's product page or refer to Leopard's Python SDK quick start guide.