miya-speechless 0.0.24

Speechless repo for sales call analysis

miya-speechless

Poetry Installation Instructions

To install the dependencies and manage the project, we use Poetry. Follow the steps below to set up your environment with Poetry.

Step 1: Install Poetry

You can install Poetry by running the official installation script:

curl -sSL https://install.python-poetry.org | python3 -

Alternatively, you can install it via pip:

pip install poetry

Step 2: Verify Poetry Installation

After installing, verify that Poetry is available by running:

poetry --version

Step 3: Install Dependencies

Once Poetry is installed, you can install the project dependencies by running the following command in the project root:

poetry install

This will create a virtual environment and install all dependencies specified in the pyproject.toml file.

For Poetry >= 2.0.0, add poetry shell plugin, based on how you have installed Poetry, choose one of:

poetry self add poetry-plugin-shell
pipx inject poetry poetry-plugin-shell
pip install poetry-plugin-shell

Step 4: Run the Project or Tests

You can now run the project or the tests using the Poetry environment:

To activate the Poetry environment:

poetry shell

To run the tests:

poetry run pytest

Step 4: Convert the model to onnx format

To convert the model to onnx format, run the following command:

poetry poetry run python export_to_onnx.py --checkpoint /path/to/checkpoint --onnx_model /path/to/onnx_model

Step 5: Run the streamlit app

Place the onnx model in the models directory.

To start the streamlit app, run the following command:

poetry run streamlit run app/app.py

To start debug mode, run:

poetry run python -m debugpy --listen 5678 -m streamlit run app/app.py

In the application, you can set the following parameters:

• overlap: Set the overlap between the transcript and the diaretization (default: 0.1)
• onset_threshold: Set the onset threshold for speaker start (default: 0.1)
• offset_threshold: Set the offset threshold for speaker stop (default: 0.1)

Step 6: Add OPENAI_API_KEY and/or setup WHISPER_CPP_MODEL

whisper_1 requires paid OpenAI subscription. An alternative is whisper.cpp

Download at least one of supported models:

#Linux
docker run -it --rm -v ./data/models:/models ghcr.io/ggerganov/whisper.cpp:main "./models/download-ggml-model.sh small /models"
# Windows
docker run -it --rm -v "$(pwd -W)/models":/models ghcr.io/ggerganov/whisper.cpp:main "./models/download-ggml-model.sh small /models"

When WHISPER_CPP_MODEL set, expect following to happen instead of calling OpenAI:

ffmpeg -i data/temp_results/uploaded_audio.mp3 -ar 16000 -ac 1 -c:a pcm_s16le data/audio/output.wav
# Linux
docker run -it --rm -v ./data/models:/models -v ./data/audio:/audios ghcr.io/ggerganov/whisper.cpp:main "./build/bin/whisper-cli -m /models/ggml-small.bin -f /audios/output.wav -ml 16 -oj -l en"
# Windows
docker run -it --rm -v "$(pwd -W)/data/models":/models -v "$(pwd -W)/data":/audios ghcr.io/ggerganov/whisper.cpp:main "./build/bin/whisper-cli -m /models/ggml-small.bin -f /audios/output.wav -ml 16 -oj -l en"