Speech analytics package for call-center allows you to estimate your operator's work. It provides a user-friendly interface using Telegram bots.
Every file goes through the following operations:
- Noise suppression
- Diarization
- Voice activity detection
- Pauses and interruptions detection
- Necessary phrases detection
- Bad words and threats detection
First, you need to clone the repository:
git clone https://github.com/DinoTheDinosaur/speech_analytics
cd speech_analytics
Create virtual environment and install necessary packages. Please note, that only Python 3.8 is supported:
python -m venv venv
python -m pip install -r requirements.txt
Create config.yaml file:
# for SR (yandex or vosk)
recognition_engine: "vosk"
# necessary files (models, corpuses, etc.)
suppressor_model_weights: "./data/model_weights.ckpt"
vosk_model: "./data/vosk"
white_list: "./data/white_list.json"
obscene_corpus: "./data/obscene_corpus.json"
threats_corpus: "./data/threats_corpus.json"
white_checklist: "./data/check_list_white.json"
black_checklist: "./data/check_list_black.json"
# for Telegram bot
bot_token: "YOUR_TELEGRAM_BOT_ACCESS_TOKEN_HERE"
# for Yandex Speech kit
bucket: "YOUR_BUCKET_HERE"
aws_key: "YOUR_AWS_KEY_HERE"
aws_key_id: "YOUR_AWS_KEY_ID_HERE"
ya_api_key: "YOUR_YANDEX_SPEECH_KIT_API_KEY"Run bot:
python run_bot.py
In case you didn't find necessary data in the repository, you can download it directly: