livestt 1.0.7

Simple and easy to use realtime speech to text

livestt

Installation

pip install livestt # this could take a while

Usage

Livestt gives access to 3 main classes/functions.

Wait for the wake word

from livestt import wait

def callback_func():
    print("Wakeword said!")

wait(callback=callback_func)

The wait function takes in these args:

callback (Callable): The function to be called when the wake word is detected.

args (tuple[any] | None): The arguments to be passed to the callback function. The default is None.

wake_word (str): The wake word that the function is waiting for. The default is "Sheila".

prob_threshold (float): The probability threshold for the wake word detection. The default is 0.5.

chunk_length_s (float): The length of the audio chunk to be processed at a time, in seconds. The default is 2.0.

stream_chunk_s (float): The length of the audio stream chunk to be processed at a time, in seconds. The default is 0.25.

debug (bool): If True, debug information will be printed. The default is True.

Raises: ValueError: If the wake word is not in the set of valid class labels.

Returns: None

Record audio

from livestt import Recorder
import time

recorder = Recorder("test.wav")

recorder.start()    # Starts recorder thread
time.sleep(5)   # Waits before ending thread
recorder.end()  # Writes recording to "test.wav"

The Recorder class when started starts a new recorder thread where it will listen to the audio until the thread is ended. Upon the thread ending, the recording will be saved to a file. The Recorder class takes these args:

chunk (int): The number of audio frames per buffer.

format (int): The sample format for the recording.

channels (int): The number of channels for the recording.

fs (int): The sample rate of the recording.

filename (str): The name of the output file where the recording will be saved. The file_ MUST currently be .wav

listening (bool): A flag indicating whether the recorder is currently recording.

Transcribe a given audio file

from livestt import transcribe

transcription = transcribe("test.wav")

for t in transcription:
    print(t.text)

The transcribe function transcribes the given audio file and outputs the transcribed text along with other information. The transcribe function takes these args:

input_file (str): The path to the audio file to be transcribed.

language (str): The language of the audio file. The default is "en" (English).

model_name (str): The name of the model to be used for transcription. The default is "tiny.en".

This function yields a tuple with the following fields:

text (str): The transcribed text.

language_probability (float): The probability of the detected language.

language (str): The detected language.

segment_end (float): The end time of the transcribed segment.

segment_start (float): The start time of the transcribed segment.

Examples

For a full example, check out the example in the file example/main.py.

Tech stack

• Pyaudio for recording audio.
• faster-whisper for transcription.
• openWakeWord for wakeword detection.

Acknowledgments

Thanks to Kolja for the inspiration. I couldn't figure out how to use his library so I made my own. Check this out here.

Contribution

Contributions are always welcome! Open an issue or make a PR. Or just contact me on discord: @a3l6

Author(s)

• @a3l6