Practical overview
When it fits
Use Diart when you are building software that needs streaming audio inference, especially when recognizing speakers or detecting voice activity is a core task.
Who it is for
- Python developers building applications that analyze live or recorded conversation audio.
- Teams that need speaker diarization or voice activity detection within an audio pipeline.
- Researchers and developers composing, benchmarking or tuning custom audio pipelines.
Documented uses
- Process a recorded conversation or microphone stream with a speaker-diarization pipeline.
- Use a voice activity detection pipeline on streaming audio.
- Connect an audio source to streaming inference and write results to an RTTM file.
- Serve a pipeline on the web using WebSockets.
Requirements & limits
The repository identifies Diart as open-source software under the MIT license. The documented package installation uses pip; access to the default pyannote models requires accepting their Hugging Face conditions and authenticating with a user access token.
- The framework documents recorded-file and microphone streaming inputs, along with Python streaming inference.
- Speaker diarization is the default CLI pipeline; voice activity detection is another documented option.
- The default pyannote models have separate usage-condition and authentication steps.
- Installation depends on specified ffmpeg, PortAudio and libsndfile versions.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Install Diart and prepare its default models
Set up the Python package and the Hugging Face access required to use its default pyannote models.
Before you start
- Python 3.10, 3.11 or 3.12
- ffmpeg below 4.4
- PortAudio 19.6.x
- libsndfile 1.2.2 or later
- A Hugging Face user access token for the default pyannote models
Steps
Install the listed system dependencies, or create and activate the supplied Conda environment.
Install the package with pip install diart.
Accept the usage conditions for the pyannote segmentation, segmentation-3.0 and embedding models.
Install huggingface-cli and log in with a user access token, or provide the token manually through the Diart CLI or API.
Expected outcome
The package is installed and the documented access steps for its default pyannote models are complete.
Guide limitations
- This covers installation and model access setup, not running or validating an audio pipeline.
Stream a recorded conversation from the command line
Start Diart's default speaker-diarization pipeline on a recorded audio file.
Before you start
- Diart installed with its documented system dependencies
- A recorded conversation in an audio file
- Hugging Face model conditions accepted and access configured for the default pyannote models
Steps
Run diart.stream followed by the path to the audio file.
Use the documented pipeline option to select VoiceActivityDetection instead of the default SpeakerDiarization pipeline when needed.
Expected outcome
Diart processes the supplied recording using the selected documented pipeline.
Guide limitations
- The README gives a command-line example but does not specify an output format for this command.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 14 October 2026
- Original website title
- juanmc2005/diart · Official homepage title · retained in its source language