Cheetah Binding for Python

Cheetah Speech-to-Text Engine

Made in Vancouver, Canada by Picovoice

Cheetah is an on-device streaming speech-to-text engine. Cheetah is:

• Private; All voice processing runs locally.
• Accurate
• Compact and Computationally-Efficient
• Cross-Platform:
  • Linux (x86_64), macOS (x86_64, arm64), and Windows (x86_64, arm64)
  • Android and iOS
  • Chrome, Safari, Firefox, and Edge
  • Raspberry Pi (3, 4, 5)

Compatibility

• Python 3.9+
• Runs on Linux (x86_64), macOS (x86_64, arm64), Windows (x86_64, arm64), and Raspberry Pi (3, 4, 5).

Installation

pip3 install pvcheetah

AccessKey

Cheetah requires a valid Picovoice AccessKey at initialization. AccessKey acts as your credentials when using Cheetah SDKs. You can get your AccessKey for free. Make sure to keep your AccessKey secret. Signup or Login to Picovoice Console to get your AccessKey.

Usage

Basic Transcription

Create an instance of the engine and transcribe audio:

import pvcheetah

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

def get_next_audio_frame():
    pass

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

Replace ${ACCESS_KEY} with yours obtained from Picovoice Console. When done be sure to explicitly release the resources using handle.delete().

Annotated Transcription

Create an instance of the engine and get the audio transcription and word-level metadata:

import pvcheetah

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

def get_next_audio_frame():
    pass

while True:
    partial_output = handle.process_annotated(get_next_audio_frame())
    partial_transcript = partial_output.transcript
    partial_words = partial_output.words
    is_endpoint = partial_output.is_endpoint

    if is_endpoint:
        final_output = handle.flush_annotated()
        final_transcript = final_output.transcript
        final_words = final_output.words

Replace ${ACCESS_KEY} with yours obtained from Picovoice Console. When done be sure to explicitly release the resources using handle.delete().

Language Model

The Cheetah Python SDK comes preloaded with a default English language model (.pv file). Default models for other supported languages can be found in lib/common.

Create custom language models using the Picovoice Console. Here you can train language models with custom vocabulary and boost words in the existing vocabulary.

Pass in the .pv file via the model_path argument:

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

Train Models over API

You can train models over API without going to the console:

train_model_from_words(
        "${ACCESS_KEY}",                                                   # AccessKey obtained from Picovoice Console (https://console.picovoice.ai/).
        "${OUTPUT_PATH}",                                                  # Path to save the newly trained model.
        "${LANGUAGE}",                                                     # Two-character language code.
        {"${NEW_WORD}": set(["${PRONUNCIATION1}", "${PRONUNCIATION2}"])},  # New words with optional custom pronunciation to add to the model.
        set(["${BOOST_WORD1}", "${BOOST_WORD2}"]))                         # Boost words.

(or)

train_model_from_yaml(
        "${ACCESS_KEY}",   # AccessKey obtained from Picovoice Console (https://console.picovoice.ai/).
        "${OUTPUT_PATH}",  # Path to save the newly trained model.
        "${LANGUAGE}",     # Two-character language code.
        "${YAML_PATH}")    # Path to YAML configuration file.

Check Cheetah Model API docs for a list of supported languages.

Demos

pvcheetahdemo provides command-line utilities for processing audio using Cheetah.