faststt 0.1.0

Hybrid Speech-to-Text using Faster-Whisper and SpeechRecognition

FastSTT

FastSTT is a lightweight, hybrid Speech-to-Text Python library with real-time microphone and audio file transcription. Ideal for virtual assistants, chatbots, and low-latency voice input applications.

Features

• Offline transcription with low latency
• Real-time microphone transcription
• Audio file transcription (WAV, MP3, M4A, etc.)
• Optional timestamps for segments
• Automatic or forced language detection
• Lightweight and easy to use
• Compatible with Python 3.10+

Installation

You can install FastSTT via PyPI:

pip install faststt

Note: Microphone support requires PyAudio installed. On some systems, you may need system dependencies:

Ubuntu/Debian:

sudo apt-get install portaudio19-dev

Windows Use the prebuilt PyAudio wheel for your Python version.

Usage

1. Transcribe from microphone

from faststt import FastSTT
import logging

logging.basicConfig(level=logging.INFO)

stt = FastSTT(model_size="base", device="cpu")

print("Speak something...")
text = stt.listen_and_transcribe(timeout=5, phrase_time_limit=10)

print("Transcription:", text)

2. Transcribe from microphone in real-time

from faststt import FastSTT
import logging

logging.basicConfig(level=logging.INFO)

stt = FastSTT(model_size="base", device="cpu")

print("Listening continuously. Press Ctrl+C to stop.\n")

try:
    while True:
        text = stt.listen_and_transcribe(timeout=3, phrase_time_limit=5)
        if text and isinstance(text, dict):
            print("Transcription:", text["text"])
        else:
            print("...no speech detected.")
except KeyboardInterrupt:
    print("\nStopped by user.")

3. Transcribe from audio file

from faststt import FastSTT

stt = FastSTT(model_size="base", device="cpu", compute_type="int8")

file_path = "audio_file.wav"
result = stt.listen_from_file(file_path, with_timestamps=True)

if result:
    print("Transcription with timestamps:")
    for seg in result:
        print(f"[{seg['start']:.2f}s - {seg['end']:.2f}s]: {seg['text']}")
else:
    print("Transcription failed.")

License

This project is licensed under the MIT License.