NeuTTS-2E: Optimized for Qualcomm Devices
NeuTTS-2E is a compact on-device emotional text-to-speech language model from Neuphonic. It pairs a 236M-parameter Qwen3 causal LM backbone with the NeuCodec 50Hz neural audio codec, and produces natural-sounding 24kHz English speech for four fixed speakers across seven emotions. There is no voice cloning, as the four speakers ship as pre-encoded reference codes, so no reference audio is read and the codec's encoder is never called. The backbone exports as two graphs (a 128-token prefill and a single-token decode) linked into one weight-shared context binary; prompt assembly, emotion conditioning, sampling and NeuCodec decode run on CPU in the app.
This is based on the implementation of NeuTTS-2E 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 |
|---|---|---|---|---|
| QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | QAIRT 2.51 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.51 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | QAIRT 2.51 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 1 Mobile | QAIRT 2.51 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-8275 | QAIRT 2.51 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | QAIRT 2.51 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-9075 | QAIRT 2.51 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | QAIRT 2.51 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Extreme Gen 6 Mobile | QAIRT 2.51 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 6 Mobile | QAIRT 2.51 | Download |
For more device-specific assets and performance metrics, visit NeuTTS-2E 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 NeuTTS-2E on GitHub for usage instructions.
Model Details
Model Type: Model_use_case.audio_generation
Model Stats:
- Model checkpoint: neuphonic/neutts-2e
- Model size (float): 671 MB
- Number of parameters: 236M
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| NeuTTS-2E (prompt_ar128_cl2048) | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 39.73 ms | 0 - 11 MB | NPU |
| NeuTTS-2E (prompt_ar128_cl2048) | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | 49.446 ms | 0 - 9 MB | NPU |
| NeuTTS-2E (prompt_ar128_cl2048) | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 64.967 ms | 1 - 1 MB | NPU |
| NeuTTS-2E (prompt_ar128_cl2048) | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 1 Mobile | 88.469 ms | 0 - 11 MB | NPU |
| NeuTTS-2E (prompt_ar128_cl2048) | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-8275 | 86.199 ms | 0 - 123 MB | NPU |
| NeuTTS-2E (prompt_ar128_cl2048) | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 72.263 ms | 0 - 3 MB | NPU |
| NeuTTS-2E (prompt_ar128_cl2048) | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8450 | 88.469 ms | 0 - 11 MB | NPU |
| NeuTTS-2E (prompt_ar128_cl2048) | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-9075 | 85.129 ms | 0 - 122 MB | NPU |
| NeuTTS-2E (prompt_ar128_cl2048) | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-X7181 | 64.967 ms | 1 - 1 MB | NPU |
| NeuTTS-2E (prompt_ar128_cl2048) | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ Q-8750 | 49.446 ms | 0 - 9 MB | NPU |
| NeuTTS-2E (prompt_ar128_cl2048) | QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | 82.951 ms | 0 - 7 MB | NPU |
| NeuTTS-2E (prompt_ar128_cl2048) | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Extreme Gen 6 Mobile | 36.579 ms | 7 - 16 MB | NPU |
| NeuTTS-2E (prompt_ar128_cl2048) | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 6 Mobile | 43.686 ms | 0 - 9 MB | NPU |
| NeuTTS-2E (token_ar1_cl2048) | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 17.603 ms | 0 - 9 MB | NPU |
| NeuTTS-2E (token_ar1_cl2048) | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | 21.364 ms | 42 - 50 MB | NPU |
| NeuTTS-2E (token_ar1_cl2048) | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 24.901 ms | 1 - 1 MB | NPU |
| NeuTTS-2E (token_ar1_cl2048) | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 1 Mobile | 36.844 ms | 1 - 11 MB | NPU |
| NeuTTS-2E (token_ar1_cl2048) | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-8275 | 38.121 ms | 0 - 4 MB | NPU |
| NeuTTS-2E (token_ar1_cl2048) | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 30.682 ms | 0 - 2 MB | NPU |
| NeuTTS-2E (token_ar1_cl2048) | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8450 | 36.844 ms | 1 - 11 MB | NPU |
| NeuTTS-2E (token_ar1_cl2048) | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-9075 | 33.636 ms | 0 - 4 MB | NPU |
| NeuTTS-2E (token_ar1_cl2048) | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-X7181 | 24.901 ms | 1 - 1 MB | NPU |
| NeuTTS-2E (token_ar1_cl2048) | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ Q-8750 | 21.364 ms | 42 - 50 MB | NPU |
| NeuTTS-2E (token_ar1_cl2048) | QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | 34.599 ms | 0 - 6 MB | NPU |
| NeuTTS-2E (token_ar1_cl2048) | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Extreme Gen 6 Mobile | 17.294 ms | 0 - 9 MB | NPU |
| NeuTTS-2E (token_ar1_cl2048) | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 6 Mobile | 18.136 ms | 0 - 9 MB | NPU |
License
- The license for the original implementation of NeuTTS-2E 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.
