MeloTTS-ZH: Optimized for Qualcomm Devices
MeloTTS is a high-quality multi-lingual text-to-speech library for English, Chinese and Spanish language.
This is based on the implementation of MeloTTS-ZH 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.
Deploying MeloTTS-ZH on-device
This model is compatible with the Qualcomm Voice AI SDK. Download the SDK from the Qualcomm Package Manager to deploy this model on-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® X2 Elite | QAIRT 2.45 | Download |
| VOICE_AI | mixed_with_float | Snapdragon® X Elite | QAIRT 2.45 | Download |
| VOICE_AI | mixed_with_float | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.45 | Download |
| VOICE_AI | mixed_with_float | Qualcomm® Dragonwing™ IQ-8275 | QAIRT 2.45 | Download |
| VOICE_AI | mixed_with_float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | QAIRT 2.45 | Download |
| VOICE_AI | mixed_with_float | Qualcomm® SA8775P | QAIRT 2.45 | Download |
| VOICE_AI | mixed_with_float | Qualcomm® Dragonwing™ IQ-9075 | QAIRT 2.45 | Download |
| VOICE_AI | mixed_with_float | Qualcomm® SA7255P | QAIRT 2.45 | Download |
| VOICE_AI | mixed_with_float | Snapdragon® 8 Elite Mobile | QAIRT 2.45 | Download |
| VOICE_AI | mixed_with_float | Snapdragon® 8 Elite Gen 5 Mobile | QAIRT 2.45 | Download |
For more device-specific assets and performance metrics, visit MeloTTS-ZH 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 MeloTTS-ZH on GitHub for usage instructions.
Model Details
Model Type: Model_use_case.audio_generation
Model Stats:
- Max decoded sequence length: 512 tokens
- Model checkpoint: myshell-ai/MeloTTS-Chinese
- Model size (bert_wrapper) (float): 581 MB
- Model size (decoder) (float): 55.5 MB
- Model size (encoder) (float): 31.9 MB
- Model size (flow) (float): 76.9 MB
- Number of parameters (bert_wrapper): 152M
- Number of parameters (decoder): 14.5M
- Number of parameters (encoder): 8.34M
- Number of parameters (flow): 20.1M
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| bert_wrapper | VOICE_AI | mixed_with_float | Snapdragon® X2 Elite | 3.268 ms | 0 - 0 MB | NPU |
| bert_wrapper | VOICE_AI | mixed_with_float | Snapdragon® X Elite | 7.753 ms | 0 - 0 MB | NPU |
| bert_wrapper | VOICE_AI | mixed_with_float | Snapdragon® 8 Gen 3 Mobile | 5.067 ms | 0 - 7 MB | NPU |
| bert_wrapper | VOICE_AI | mixed_with_float | Qualcomm® Dragonwing™ IQ-8275 | 30.616 ms | 0 - 9 MB | NPU |
| bert_wrapper | VOICE_AI | mixed_with_float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 7.107 ms | 0 - 1 MB | NPU |
| bert_wrapper | VOICE_AI | mixed_with_float | Qualcomm® SA8775P | 9.284 ms | 0 - 10 MB | NPU |
| bert_wrapper | VOICE_AI | mixed_with_float | Qualcomm® SA8650P | 9.284 ms | 0 - 10 MB | NPU |
| bert_wrapper | VOICE_AI | mixed_with_float | Qualcomm® SA8255P | 9.284 ms | 0 - 10 MB | NPU |
| bert_wrapper | VOICE_AI | mixed_with_float | Qualcomm® Dragonwing™ IQ-9075 | 9.012 ms | 0 - 2 MB | NPU |
| bert_wrapper | VOICE_AI | mixed_with_float | Qualcomm® Dragonwing™ IQ-X7181 | 7.753 ms | 0 - 0 MB | NPU |
| bert_wrapper | VOICE_AI | mixed_with_float | Qualcomm® Dragonwing™ Q-8750 | 3.34 ms | 0 - 13 MB | NPU |
| bert_wrapper | VOICE_AI | mixed_with_float | Qualcomm® SA7255P | 30.616 ms | 0 - 9 MB | NPU |
| bert_wrapper | VOICE_AI | mixed_with_float | Snapdragon® 8 Elite Mobile | 3.34 ms | 0 - 13 MB | NPU |
| bert_wrapper | VOICE_AI | mixed_with_float | Snapdragon® 8 Elite Gen 5 Mobile | 2.691 ms | 0 - 8 MB | NPU |
| decoder | VOICE_AI | mixed_with_float | Snapdragon® X2 Elite | 40.451 ms | 0 - 0 MB | NPU |
| decoder | VOICE_AI | mixed_with_float | Snapdragon® X Elite | 82.728 ms | 0 - 0 MB | NPU |
| decoder | VOICE_AI | mixed_with_float | Snapdragon® 8 Gen 3 Mobile | 60.613 ms | 0 - 7 MB | NPU |
| decoder | VOICE_AI | mixed_with_float | Qualcomm® Dragonwing™ IQ-8275 | 134.556 ms | 0 - 10 MB | NPU |
| decoder | VOICE_AI | mixed_with_float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 87.247 ms | 1 - 2 MB | NPU |
| decoder | VOICE_AI | mixed_with_float | Qualcomm® SA8775P | 83.852 ms | 0 - 10 MB | NPU |
| decoder | VOICE_AI | mixed_with_float | Qualcomm® SA8650P | 83.852 ms | 0 - 10 MB | NPU |
| decoder | VOICE_AI | mixed_with_float | Qualcomm® SA8255P | 83.852 ms | 0 - 10 MB | NPU |
| decoder | VOICE_AI | mixed_with_float | Qualcomm® Dragonwing™ IQ-9075 | 94.677 ms | 0 - 2 MB | NPU |
| decoder | VOICE_AI | mixed_with_float | Qualcomm® Dragonwing™ IQ-X7181 | 82.728 ms | 0 - 0 MB | NPU |
| decoder | VOICE_AI | mixed_with_float | Qualcomm® Dragonwing™ Q-8750 | 48.08 ms | 0 - 13 MB | NPU |
| decoder | VOICE_AI | mixed_with_float | Qualcomm® SA7255P | 134.556 ms | 0 - 10 MB | NPU |
| decoder | VOICE_AI | mixed_with_float | Snapdragon® 8 Elite Mobile | 48.08 ms | 0 - 13 MB | NPU |
| decoder | VOICE_AI | mixed_with_float | Snapdragon® 8 Elite Gen 5 Mobile | 42.575 ms | 0 - 8 MB | NPU |
| encoder | VOICE_AI | mixed_with_float | Snapdragon® X2 Elite | 25.283 ms | 4 - 4 MB | NPU |
| encoder | VOICE_AI | mixed_with_float | Snapdragon® X Elite | 40.169 ms | 4 - 4 MB | NPU |
| encoder | VOICE_AI | mixed_with_float | Snapdragon® 8 Gen 3 Mobile | 30.183 ms | 4 - 12 MB | NPU |
| encoder | VOICE_AI | mixed_with_float | Qualcomm® Dragonwing™ IQ-8275 | 66.311 ms | 2 - 11 MB | NPU |
| encoder | VOICE_AI | mixed_with_float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 40.549 ms | 4 - 6 MB | NPU |
| encoder | VOICE_AI | mixed_with_float | Qualcomm® SA8775P | 43.166 ms | 2 - 11 MB | NPU |
| encoder | VOICE_AI | mixed_with_float | Qualcomm® SA8650P | 43.166 ms | 2 - 11 MB | NPU |
| encoder | VOICE_AI | mixed_with_float | Qualcomm® SA8255P | 43.166 ms | 2 - 11 MB | NPU |
| encoder | VOICE_AI | mixed_with_float | Qualcomm® Dragonwing™ IQ-9075 | 42.726 ms | 4 - 9 MB | NPU |
| encoder | VOICE_AI | mixed_with_float | Qualcomm® Dragonwing™ IQ-X7181 | 40.169 ms | 4 - 4 MB | NPU |
| encoder | VOICE_AI | mixed_with_float | Qualcomm® Dragonwing™ Q-8750 | 25.45 ms | 2 - 12 MB | NPU |
| encoder | VOICE_AI | mixed_with_float | Qualcomm® SA7255P | 66.311 ms | 2 - 11 MB | NPU |
| encoder | VOICE_AI | mixed_with_float | Snapdragon® 8 Elite Mobile | 25.45 ms | 2 - 12 MB | NPU |
| encoder | VOICE_AI | mixed_with_float | Snapdragon® 8 Elite Gen 5 Mobile | 23.867 ms | 2 - 10 MB | NPU |
| flow | VOICE_AI | mixed_with_float | Snapdragon® X2 Elite | 67.358 ms | 2 - 2 MB | NPU |
| flow | VOICE_AI | mixed_with_float | Snapdragon® X Elite | 129.204 ms | 2 - 2 MB | NPU |
| flow | VOICE_AI | mixed_with_float | Snapdragon® 8 Gen 3 Mobile | 99.141 ms | 4 - 12 MB | NPU |
| flow | VOICE_AI | mixed_with_float | Qualcomm® Dragonwing™ IQ-8275 | 245.546 ms | 2 - 11 MB | NPU |
| flow | VOICE_AI | mixed_with_float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 132.415 ms | 3 - 4 MB | NPU |
| flow | VOICE_AI | mixed_with_float | Qualcomm® SA8775P | 128.681 ms | 2 - 12 MB | NPU |
| flow | VOICE_AI | mixed_with_float | Qualcomm® SA8650P | 128.681 ms | 2 - 12 MB | NPU |
| flow | VOICE_AI | mixed_with_float | Qualcomm® SA8255P | 128.681 ms | 2 - 12 MB | NPU |
| flow | VOICE_AI | mixed_with_float | Qualcomm® Dragonwing™ IQ-9075 | 127.428 ms | 4 - 7 MB | NPU |
| flow | VOICE_AI | mixed_with_float | Qualcomm® Dragonwing™ IQ-X7181 | 129.204 ms | 2 - 2 MB | NPU |
| flow | VOICE_AI | mixed_with_float | Qualcomm® Dragonwing™ Q-8750 | 81.895 ms | 2 - 12 MB | NPU |
| flow | VOICE_AI | mixed_with_float | Qualcomm® SA7255P | 245.546 ms | 2 - 11 MB | NPU |
| flow | VOICE_AI | mixed_with_float | Snapdragon® 8 Elite Mobile | 81.895 ms | 2 - 12 MB | NPU |
| flow | VOICE_AI | mixed_with_float | Snapdragon® 8 Elite Gen 5 Mobile | 71.056 ms | 2 - 11 MB | NPU |
License
- The license for the original implementation of MeloTTS-ZH can be found here.
References
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.
