|
|
--- |
|
|
license: apache-2.0 |
|
|
language: |
|
|
- en |
|
|
metrics: |
|
|
- accuracy |
|
|
--- |
|
|
# Lattice-1-Alpha |
|
|
|
|
|
## Model Description |
|
|
|
|
|
Lattice-1-Alpha is an advanced forced alignment model designed for precise word-level alignment between audio and text. This model provides state-of-the-art performance in aligning speech with its corresponding word transcription, enabling accurate time-stamping of words in spoken language. |
|
|
|
|
|
## Key Features |
|
|
|
|
|
- **High-Precision Alignment**: Achieves accurate word-level forced alignment with millisecond precision |
|
|
- **Language Support**: Currently supports English only. The upcoming **Lattice-1** release will support English, Chinese, and mixed English-Chinese content. |
|
|
- **Efficient Processing**: Optimized for fast inference with ONNX runtime support |
|
|
- **Flexible Integration**: Easy to integrate into various speech processing pipelines |
|
|
|
|
|
## Quick Start with Python SDK |
|
|
|
|
|
The easiest way to use Lattice-1-Alpha is through the **[Lattifai Python SDK](https://github.com/lattifai/lattifai-python)**. |
|
|
|
|
|
### Installation |
|
|
|
|
|
```bash |
|
|
pip install install-k2 |
|
|
install-k2 # Install k2 dependencies |
|
|
|
|
|
pip install lattifai |
|
|
``` |
|
|
|
|
|
### Basic Usage |
|
|
|
|
|
```shell |
|
|
# Align audio with subtitle |
|
|
lattifai align audio.wav subtitle.srt output.srt |
|
|
``` |
|
|
|
|
|
For more detailed usage instructions and API documentation, please visit: |
|
|
**[Lattifai Python SDK Documentation](https://github.com/lattifai/lattifai-python)** |
|
|
|
|
|
|
|
|
## Applications |
|
|
|
|
|
- **Speech Data Processing**: Creating high-quality training data for ASR, TTS, and Speech LLM systems |
|
|
- **Pronunciation Assessment**: Evaluating learner pronunciation accuracy |
|
|
- **Audio-Text Synchronization**: Subtitle generation and synchronization |
|
|
- **Voice Conversion**: Time-aligned feature extraction for voice transformation |
|
|
|
|
|
## Performance |
|
|
|
|
|
Lattice-1-Alpha has been evaluated on standard benchmarks and demonstrates: |
|
|
- High accuracy in word boundary detection |
|
|
- Robust performance across different speaking styles |
|
|
- Fast inference speed suitable for real-time applications |
|
|
|
|
|
## Citation |
|
|
|
|
|
If you use Lattice-1-Alpha in your research, please cite: |
|
|
|
|
|
```bibtex |
|
|
@misc{lattice1alpha, |
|
|
title={Lattice-1-Alpha: High-Precision Word-Level Forced Alignment}, |
|
|
author={Lattifai Team}, |
|
|
year={2025}, |
|
|
publisher={Hugging Face}, |
|
|
howpublished={\url{https://huggingface.co/lattifai/Lattice-1-Alpha}} |
|
|
} |
|
|
``` |
|
|
|
|
|
## License |
|
|
|
|
|
Please refer to the license file for usage terms and conditions. |
|
|
|
|
|
## Support |
|
|
|
|
|
For issues, questions, or feature requests, please visit: |
|
|
- GitHub: [https://github.com/lattifai/lattifai-python](https://github.com/lattifai/lattifai-python) |
|
|
- Documentation: [Lattifai Python SDK Documentation](https://github.com/lattifai/lattifai-python?tab=readme-ov-file#quick-start) |
|
|
|
|
|
## Acknowledgments |
|
|
|
|
|
This model is developed and maintained by the Lattifai team. |
