README.md

---
library_name: pytorch
license: mit
tags:
- real_time
- android
pipeline_tag: automatic-speech-recognition

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/funasr_conformer_en/web-assets/model_demo.png)

# FunASR-Conformer-EN: Optimized for Qualcomm Devices

FunASR Conformer-EN is a large-scale English ASR model from Alibaba DAMO Academy, trained on ~50,000 hours of English speech. It uses a 32-block Conformer encoder(512 hidden dim, 16 attention heads) with 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 FunASR-Conformer-EN found [here](https://huggingface.co/funasr/conformer-en).
This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.61.0/src/qai_hub_models/models/funasr_conformer_en) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).

Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) 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 |
|---|---|---|---|---|
| ONNX | float | Universal | QAIRT 2.45, ONNX Runtime 1.27.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/funasr_conformer_en/releases/v0.61.0/funasr_conformer_en-onnx-float.zip)
| QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/funasr_conformer_en/releases/v0.61.0/funasr_conformer_en-qnn_dlc-float.zip)
| TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/funasr_conformer_en/releases/v0.61.0/funasr_conformer_en-tflite-float.zip)

For more device-specific assets and performance metrics, visit **[FunASR-Conformer-EN on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/funasr_conformer_en)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.61.0/src/qai_hub_models/models/funasr_conformer_en) 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 [FunASR-Conformer-EN on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.61.0/src/qai_hub_models/models/funasr_conformer_en) for usage instructions.

## Model Details

**Model Type:** Model_use_case.speech_recognition

**Model Stats:**
- Input resolution: 1x160000 (10s at 16kHz)
- Model checkpoint: funasr/conformer-en
- Model size (float): ~840 MB
- Number of parameters: 220M

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| FunASR-Conformer-EN | ONNX | float | Snapdragon® X2 Elite | 24.852 ms | 9 - 9 MB | NPU
| FunASR-Conformer-EN | ONNX | float | Snapdragon® X Elite | 50.732 ms | 326 - 326 MB | NPU
| FunASR-Conformer-EN | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 33.644 ms | 7 - 1060 MB | NPU
| FunASR-Conformer-EN | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 69.257 ms | 6 - 1029 MB | NPU
| FunASR-Conformer-EN | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 54.153 ms | 6 - 9 MB | NPU
| FunASR-Conformer-EN | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 47.515 ms | 0 - 387 MB | NPU
| FunASR-Conformer-EN | ONNX | float | Qualcomm® QCS8450 | 69.257 ms | 6 - 1029 MB | NPU
| FunASR-Conformer-EN | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 54.526 ms | 6 - 9 MB | NPU
| FunASR-Conformer-EN | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 50.732 ms | 326 - 326 MB | NPU
| FunASR-Conformer-EN | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 24.062 ms | 7 - 722 MB | NPU
| FunASR-Conformer-EN | ONNX | float | Snapdragon® 8 Elite Mobile | 24.062 ms | 7 - 722 MB | NPU
| FunASR-Conformer-EN | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 19.34 ms | 6 - 704 MB | NPU
| FunASR-Conformer-EN | QNN_DLC | float | Snapdragon® X2 Elite | 23.738 ms | 0 - 0 MB | NPU
| FunASR-Conformer-EN | QNN_DLC | float | Snapdragon® X Elite | 47.954 ms | 0 - 0 MB | NPU
| FunASR-Conformer-EN | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 31.812 ms | 0 - 953 MB | NPU
| FunASR-Conformer-EN | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 65.029 ms | 1 - 922 MB | NPU
| FunASR-Conformer-EN | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 112.899 ms | 1 - 654 MB | NPU
| FunASR-Conformer-EN | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 44.742 ms | 1 - 756 MB | NPU
| FunASR-Conformer-EN | QNN_DLC | float | Qualcomm® SA8775P | 51.223 ms | 1 - 653 MB | NPU
| FunASR-Conformer-EN | QNN_DLC | float | Qualcomm® SA8650P | 51.223 ms | 1 - 653 MB | NPU
| FunASR-Conformer-EN | QNN_DLC | float | Qualcomm® SA8255P | 51.223 ms | 1 - 653 MB | NPU
| FunASR-Conformer-EN | QNN_DLC | float | Qualcomm® QCS8450 | 65.029 ms | 1 - 922 MB | NPU
| FunASR-Conformer-EN | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 67.266 ms | 2 - 7 MB | NPU
| FunASR-Conformer-EN | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 47.954 ms | 0 - 0 MB | NPU
| FunASR-Conformer-EN | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 23.406 ms | 0 - 652 MB | NPU
| FunASR-Conformer-EN | QNN_DLC | float | Qualcomm® SA7255P | 112.899 ms | 1 - 654 MB | NPU
| FunASR-Conformer-EN | QNN_DLC | float | Qualcomm® SA8295P | 55.177 ms | 1 - 660 MB | NPU
| FunASR-Conformer-EN | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 23.406 ms | 0 - 652 MB | NPU
| FunASR-Conformer-EN | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 18.134 ms | 0 - 609 MB | NPU
| FunASR-Conformer-EN | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 32.051 ms | 3 - 1029 MB | NPU
| FunASR-Conformer-EN | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 65.923 ms | 4 - 984 MB | NPU
| FunASR-Conformer-EN | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 51.053 ms | 3 - 335 MB | NPU
| FunASR-Conformer-EN | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 112.966 ms | 3 - 683 MB | NPU
| FunASR-Conformer-EN | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 44.304 ms | 3 - 7 MB | NPU
| FunASR-Conformer-EN | TFLITE | float | Qualcomm® SA8775P | 51.703 ms | 3 - 684 MB | NPU
| FunASR-Conformer-EN | TFLITE | float | Qualcomm® SA8650P | 51.703 ms | 3 - 684 MB | NPU
| FunASR-Conformer-EN | TFLITE | float | Qualcomm® SA8255P | 51.703 ms | 3 - 684 MB | NPU
| FunASR-Conformer-EN | TFLITE | float | Qualcomm® QCS8450 | 65.923 ms | 4 - 984 MB | NPU
| FunASR-Conformer-EN | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 52.859 ms | 3 - 335 MB | NPU
| FunASR-Conformer-EN | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 23.368 ms | 3 - 690 MB | NPU
| FunASR-Conformer-EN | TFLITE | float | Qualcomm® SA7255P | 112.966 ms | 3 - 683 MB | NPU
| FunASR-Conformer-EN | TFLITE | float | Qualcomm® SA8295P | 55.196 ms | 3 - 689 MB | NPU
| FunASR-Conformer-EN | TFLITE | float | Snapdragon® 8 Elite Mobile | 23.368 ms | 3 - 690 MB | NPU
| FunASR-Conformer-EN | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 18.073 ms | 3 - 644 MB | NPU

## License
* The license for the original implementation of FunASR-Conformer-EN can be found
  [here](https://github.com/modelscope/FunASR?tab=MIT-1-ov-file).

## References
* [Conformer: Convolution-augmented Transformer for Speech Recognition](https://arxiv.org/abs/2005.08100)
* [Source Model Implementation](https://huggingface.co/funasr/conformer-en)

## Community
* Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
* For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).