Pyannote-Speaker-Diarization: Optimized for Qualcomm Devices

Pyannote Speaker Diarization is an open-source speaker diarization model that identifies "who spoke when" in an audio recording. It detects per-frame speaker activity and encodes speaker identity, enabling accurate multi-speaker attribution in meetings, interviews, and other multi-party conversations.

This is based on the implementation of Pyannote-Speaker-Diarization found here. This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the Qualcomm® AI Hub Models library to export with custom configurations. More details on model performance across various devices, can be found here.

Qualcomm AI Hub Models uses Qualcomm AI Hub Workbench to compile, profile, and evaluate this model. Sign up to run these models on a hosted Qualcomm® device.

Getting Started

There are two ways to deploy this model on your device:

Option 1: Download Pre-Exported Models

Below are pre-exported model assets ready for deployment.

Runtime	Precision	Chipset	SDK Versions	Download
VOICE_AI	mixed_with_float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	QAIRT 2.50	Download
VOICE_AI	mixed_with_float	Snapdragon® 8 Elite For Galaxy Mobile	QAIRT 2.50	Download
VOICE_AI	mixed_with_float	Snapdragon® X2 Elite	QAIRT 2.50	Download
VOICE_AI	mixed_with_float	Snapdragon® X Elite	QAIRT 2.50	Download
VOICE_AI	mixed_with_float	Snapdragon® 8 Gen 3 Mobile	QAIRT 2.50	Download
VOICE_AI	mixed_with_float	Snapdragon® 8 Gen 1 Mobile	QAIRT 2.50	Download
VOICE_AI	mixed_with_float	Qualcomm® Dragonwing™ IQ-8275	QAIRT 2.50	Download
VOICE_AI	mixed_with_float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	QAIRT 2.50	Download
VOICE_AI	mixed_with_float	Qualcomm® SA8775P	QAIRT 2.50	Download
VOICE_AI	mixed_with_float	Qualcomm® Dragonwing™ IQ-9075	QAIRT 2.50	Download
VOICE_AI	mixed_with_float	Qualcomm® SA7255P	QAIRT 2.50	Download
VOICE_AI	mixed_with_float	Qualcomm® SA8295P	QAIRT 2.50	Download

For more device-specific assets and performance metrics, visit Pyannote-Speaker-Diarization on Qualcomm® AI Hub.

Option 2: Export with Custom Configurations

Use the Qualcomm® AI Hub Models Python library to compile and export the model with your own:

• Custom weights (e.g., fine-tuned checkpoints)
• Custom input shapes
• Target device and runtime configurations

This option is ideal if you need to customize the model beyond the default configuration provided here.

See our repository for Pyannote-Speaker-Diarization on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.speaker_diarization

Model Stats:

• Input resolution (embedding): 32 x 498 x 80 (batch x frames x mel-bins)
• Input resolution (segmentation): 32 x 1 x 80000 (batch x channel x 5s at 16kHz)
• Model checkpoint: pyannote/speaker-diarization-3.1
• Number of speakers (segmentation output): 7 (powerset encoding for up to 3 speakers)
• Quantization: Segmentation float, Embedding W8A16 via AI Hub

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
embedding	VOICE_AI	mixed_with_float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	46.522 ms	5 - 12 MB	NPU
embedding	VOICE_AI	mixed_with_float	Snapdragon® 8 Elite For Galaxy Mobile	56.002 ms	0 - 9 MB	NPU
embedding	VOICE_AI	mixed_with_float	Snapdragon® X2 Elite	54.269 ms	5 - 5 MB	NPU
embedding	VOICE_AI	mixed_with_float	Snapdragon® X Elite	89.688 ms	5 - 5 MB	NPU
embedding	VOICE_AI	mixed_with_float	Snapdragon® 8 Gen 3 Mobile	72.801 ms	7 - 15 MB	NPU
embedding	VOICE_AI	mixed_with_float	Snapdragon® 8 Gen 1 Mobile	134.416 ms	6 - 15 MB	NPU
embedding	VOICE_AI	mixed_with_float	Qualcomm® Dragonwing™ IQ-8275	94.724 ms	5 - 13 MB	NPU
embedding	VOICE_AI	mixed_with_float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	93.453 ms	5 - 7 MB	NPU
embedding	VOICE_AI	mixed_with_float	Qualcomm® QCS8450	134.416 ms	6 - 15 MB	NPU
embedding	VOICE_AI	mixed_with_float	Qualcomm® Dragonwing™ IQ-9075	102.216 ms	5 - 12 MB	NPU
embedding	VOICE_AI	mixed_with_float	Qualcomm® Dragonwing™ IQ-X7181	89.688 ms	5 - 5 MB	NPU
embedding	VOICE_AI	mixed_with_float	Qualcomm® Dragonwing™ Q-8750	56.002 ms	0 - 9 MB	NPU
embedding	VOICE_AI	mixed_with_float	Qualcomm® SA8295P	146.078 ms	3 - 9 MB	NPU
segmentation	VOICE_AI	mixed_with_float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	117.972 ms	2 - 11 MB	NPU
segmentation	VOICE_AI	mixed_with_float	Snapdragon® 8 Elite For Galaxy Mobile	133.858 ms	1 - 13 MB	NPU
segmentation	VOICE_AI	mixed_with_float	Snapdragon® X2 Elite	114.91 ms	10 - 10 MB	NPU
segmentation	VOICE_AI	mixed_with_float	Snapdragon® X Elite	231.884 ms	10 - 10 MB	NPU
segmentation	VOICE_AI	mixed_with_float	Snapdragon® 8 Gen 3 Mobile	184.926 ms	10 - 18 MB	NPU
segmentation	VOICE_AI	mixed_with_float	Snapdragon® 8 Gen 1 Mobile	288.678 ms	9 - 19 MB	NPU
segmentation	VOICE_AI	mixed_with_float	Qualcomm® Dragonwing™ IQ-8275	248.915 ms	10 - 23 MB	NPU
segmentation	VOICE_AI	mixed_with_float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	228.562 ms	10 - 12 MB	NPU
segmentation	VOICE_AI	mixed_with_float	Qualcomm® QCS8450	288.678 ms	9 - 19 MB	NPU
segmentation	VOICE_AI	mixed_with_float	Qualcomm® Dragonwing™ IQ-9075	247.454 ms	10 - 22 MB	NPU
segmentation	VOICE_AI	mixed_with_float	Qualcomm® Dragonwing™ IQ-X7181	231.884 ms	10 - 10 MB	NPU
segmentation	VOICE_AI	mixed_with_float	Qualcomm® Dragonwing™ Q-8750	133.858 ms	1 - 13 MB	NPU
segmentation	VOICE_AI	mixed_with_float	Qualcomm® SA8295P	296.485 ms	0 - 6 MB	NPU

License

• The license for the original implementation of Pyannote-Speaker-Diarization can be found here.

References

• pyannote.audio: neural building blocks for speaker diarization
• Source Model Implementation

Community

• Join our AI Hub Slack community to collaborate, post questions and learn more about on-device AI.
• For questions or feedback please reach out to us.