|
|
--- |
|
|
license: cc-by-nc-nd-4.0 |
|
|
--- |
|
|
# Vietnamese Speech-to-Text (ASR) — ZipFormer-30M-RNNT-6000h |
|
|
|
|
|
## 🔍 Overview |
|
|
The **Vietnamese Speech-to-Text (ASR)** model is built on the **ZipFormer architecture** — an improved variant of the Conformer — featuring only **30 million parameters** yet achieving **exceptional performance** in both speed and accuracy. |
|
|
On CPU, the model can transcribe a **12-second audio clip in just 0.3 seconds**, significantly faster than most traditional ASR systems without requiring a GPU. |
|
|
|
|
|
--- |
|
|
|
|
|
## 🚀 Online Demo |
|
|
|
|
|
You can try the model directly here: |
|
|
👉 https://huggingface.co/spaces/hynt/k2-automatic-speech-recognition-demo |
|
|
|
|
|
--- |
|
|
|
|
|
## ⚙️ Model Architecture and Training strategy: |
|
|
- **Architecture:** ZipFormer |
|
|
- **Parameters:** ~30M |
|
|
- **Language:** Vietnamese |
|
|
- **Loss Function:** RNN-Transducer (RNNT Loss) |
|
|
- **Framework:** PyTorch + k2 |
|
|
- **Training strategy**: Carefully preprocess the data, apply an augmentation strategy based on the distribution of out-of-vocabulary (OOV) tokens and refine the transcriptions using Whisper. |
|
|
- **Optimized for:** High-speed CPU inference |
|
|
|
|
|
--- |
|
|
|
|
|
## 🧠 Training Data |
|
|
The model was trained on approximately **6000 hours of high-quality Vietnamese speech** collected from various public datasets: |
|
|
|
|
|
| Dataset | | | |
|
|
|----------|----------|----------| |
|
|
| VLSP2020 | VLSP2021 | VLSP2023-voting-pseudo-labeled | |
|
|
| VLSP2023 | FPT | VIET_BUD500 | |
|
|
| VietSpeech | FLEURS | VietMed_Labeled | |
|
|
| Sub-GigaSpeech2-Vi | ViVoice | Sub-PhoAudioBook | |
|
|
|
|
|
--- |
|
|
|
|
|
## 🧪 Evaluation Results |
|
|
|
|
|
| **Dataset** | **ZipFormer-30M-6000h** | **ChunkFormer-110M-3000h** | **PhoWhisper-Large-1.5B-800h** | **VietASR-ZipFormer-68M-70.000h** | |
|
|
|--------------|--------------------------|-----------------------------|--------------------------------|---------------------------------| |
|
|
| **VLSP2020-Test-T1** | **12.29** | 14.09 | 13.75 | 14.45 | |
|
|
| **VLSP2023-PublicTest** | **10.40** | 16.15 | 16.83 | 14.70 | |
|
|
| **VLSP2023-PrivateTest** | **11.10** | 17.12 | 17.10 | 15.07 | |
|
|
| **VLSP2025-PublicTest** | **7.97** | 15.55 | 16.14 | 13.55 | |
|
|
| **VLSP2025-PrivateTest** | **8.10** | 16.07 | 16.31 | 13.97 | |
|
|
| **GigaSpeech2-Test** | 7.56 | 10.35 | 10.00 | **6.88** | |
|
|
|
|
|
> Lower is better (WER %) |
|
|
|
|
|
--- |
|
|
|
|
|
## 🏆 Achievements |
|
|
By training this model architecture on 4,000 hours of data, I **won First Place** in the **Vietnamese Language Speech Processing (VLSP)** competition **2025**. |
|
|
Comprehensive details about **training data**, **optimization strategies**, **architecture improvements**, and **evaluation methodologies** are available in the paper below: |
|
|
|
|
|
👉 [Read the full paper on Overleaf](https://www.overleaf.com/read/wjntrgchhbgv#48aa25) |
|
|
|
|
|
--- |
|
|
|
|
|
## ⚡ Inference Speed |
|
|
|
|
|
| **Device** | **Audio Length** | **Inference Time** | |
|
|
|-------------|------------------|--------------------| |
|
|
| CPU (Hugging Face Basic) | 12 seconds | **0.3 s** | |
|
|
| GPU (RTX 3090) | 12 seconds | **< 0.1 s** | |
|
|
|
|
|
--- |
|
|
|
|
|
## ⚙️ How to Run This Model |
|
|
|
|
|
Please refer to the following guides for instructions on how to run and deploy this model: |
|
|
- **For Torch JIT Script:** [https://k2-fsa.github.io/sherpa/](https://k2-fsa.github.io/sherpa/) |
|
|
- **For ONNX:** [https://k2-fsa.github.io/sherpa/onnx/](https://k2-fsa.github.io/sherpa/onnx/) |
|
|
|
|
|
## 💬 Summary |
|
|
The **ZipFormer-30M-RNNT-6000h** model demonstrates that a lightweight architecture can still achieve state-of-the-art accuracy for Vietnamese ASR. |
|
|
It is designed for **fast deployment on CPU-based systems**, making it ideal for **real-time speech recognition**, **callbots**, and **embedded speech interfaces**. |
|
|
|
|
|
--- |
