wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations
Paper • 2006.11477 • Published • 10
Wav2Vec2-Conformer-Large-960h: Optimized for Qualcomm Devices
Facebook Wav2Vec2-Conformer-Large is an English ASR model fine-tuned on 960 hours of LibriSpeech. It uses a Conformer encoder with relative-position attention (24 layers, 1024 hidden dim, 16 heads) and a CTC head for decoding. The model accepts raw 16kHz audio and outputs transcribed text via CTC greedy decoding.
This is based on the implementation of Wav2Vec2-Conformer-Large-960h 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 |
|---|---|---|---|---|
| PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 1 Mobile | QAIRT 2.50 | Download |
| PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | w8a16 | Snapdragon® X2 Elite | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | w8a16 | Snapdragon® X Elite | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-8275 | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | w8a16 | Qualcomm® Dragonwing™ Q-6690 | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 1 Mobile | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-8275 | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-9075 | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® SA7255P | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | w8a16 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | w8a16 | Snapdragon® X2 Elite | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | w8a16 | Snapdragon® X Elite | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | w8a16 | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | w8a16 | Qualcomm® Dragonwing™ IQ-8275 | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | w8a16 | Qualcomm® SA8775P | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | w8a16 | Qualcomm® Dragonwing™ Q-6690 | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | w8a16 | Qualcomm® SA7255P | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | w8a16 | Snapdragon® 7 Gen 4 Mobile | QAIRT 2.50 | Download |
For more device-specific assets and performance metrics, visit Wav2Vec2-Conformer-Large-960h 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 Wav2Vec2-Conformer-Large-960h on GitHub for usage instructions.
Model Details
Model Type: Model_use_case.speech_recognition
Model Stats:
- Input resolution: 1x160000 (10s at 16kHz)
- Model checkpoint: facebook/wav2vec2-conformer-rel-pos-large-960h-ft
- Model size (float): ~2.4 GB
- Number of parameters: ~600M
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| Wav2Vec2-Conformer-Large-960h | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 693.553 ms | 1 - 11 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 768.25 ms | 1 - 12 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | 652.021 ms | 2 - 2 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | 959.284 ms | 1241 - 1241 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | 900.54 ms | 1 - 11 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 1167.742 ms | 0 - 5 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1053.332 ms | 0 - 1282 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 1161.021 ms | 0 - 4 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 959.284 ms | 1241 - 1241 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 768.25 ms | 1 - 12 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | PRECOMPILED_QNN_ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 747.019 ms | 1 - 14 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | PRECOMPILED_QNN_ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 821.649 ms | 1 - 12 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | PRECOMPILED_QNN_ONNX | w8a16 | Snapdragon® X2 Elite | 589.036 ms | 1 - 1 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | PRECOMPILED_QNN_ONNX | w8a16 | Snapdragon® X Elite | 1022.214 ms | 651 - 651 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | PRECOMPILED_QNN_ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 1081.808 ms | 1 - 13 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | PRECOMPILED_QNN_ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-8275 | 1121.265 ms | 0 - 4 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | PRECOMPILED_QNN_ONNX | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1242.677 ms | 0 - 770 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | PRECOMPILED_QNN_ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 1090.738 ms | 0 - 4 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | PRECOMPILED_QNN_ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 1022.214 ms | 651 - 651 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | PRECOMPILED_QNN_ONNX | w8a16 | Qualcomm® Dragonwing™ Q-6690 | 10306.401 ms | 3 - 14 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | PRECOMPILED_QNN_ONNX | w8a16 | Qualcomm® Dragonwing™ Q-7790 | 1292.306 ms | 0 - 11 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | PRECOMPILED_QNN_ONNX | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 821.649 ms | 1 - 12 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | PRECOMPILED_QNN_ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 1292.306 ms | 0 - 11 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 671.698 ms | 1 - 9 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | 767.872 ms | 1 - 10 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | 649.563 ms | 1 - 1 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 957.022 ms | 1 - 1 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | 920.387 ms | 1 - 9 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-8275 | 1167.529 ms | 1 - 4 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1052.243 ms | 2 - 3 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-9075 | 1161.157 ms | 1 - 3 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-X7181 | 957.022 ms | 1 - 1 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ Q-8750 | 767.872 ms | 1 - 10 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | 1361.581 ms | 4 - 9 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | w8a16 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 707.657 ms | 3 - 11 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 792.296 ms | 1 - 13 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | w8a16 | Snapdragon® X2 Elite | 586.51 ms | 1 - 1 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | w8a16 | Snapdragon® X Elite | 1017.919 ms | 1 - 1 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | w8a16 | Snapdragon® 8 Gen 3 Mobile | 1002.743 ms | 1 - 8 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | w8a16 | Qualcomm® Dragonwing™ IQ-8275 | 1119.974 ms | 0 - 3 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1246.819 ms | 1 - 2 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 1091.233 ms | 0 - 3 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 1017.919 ms | 1 - 1 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | w8a16 | Qualcomm® Dragonwing™ Q-6690 | 7544.425 ms | 1 - 12 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | w8a16 | Qualcomm® Dragonwing™ Q-7790 | 1288.64 ms | 0 - 6 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 792.296 ms | 1 - 13 MB | NPU |
| Wav2Vec2-Conformer-Large-960h | QNN_CONTEXT_BINARY | w8a16 | Snapdragon® 7 Gen 4 Mobile | 1288.64 ms | 0 - 6 MB | NPU |
License
- The license for the original implementation of Wav2Vec2-Conformer-Large-960h can be found here.
References
- wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations
- 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.