You can deploy this model and start earning money today!

imedemi/
sherpa-onnx-ort-zipformer-en-2023-06-26

Name: sherpa-onnx-ort-zipformer-en-2023-06-26
Availability: InStock

Views

No views yet

sherpa-onnx-ort-zipformer-en-2023-06-26

ORT-converted version of csukuangfj/sherpa-onnx-zipformer-en-2023-06-26 for faster model loading (~2.8x speedup).

Model details

Architecture: Offline zipformer transducer (65.5M params)
Training data: LibriSpeech (960h)
Format: ORT (pre-optimized ONNX Runtime format)
Quantization: int8
Files: encoder (67 MB), decoder (1.3 MB), joiner (256 KB), tokens.txt

Conversion

bash

pip install onnxruntime onnx
python -m onnxruntime.tools.convert_onnx_models_to_ort --optimization_style=Fixed .

Usage with sherpa-onnx

Use OfflineRecognizerConfig with OfflineTransducerModelConfig. The ORT format is loaded the same way as ONNX — the runtime auto-detects the format.

imedemi/sherpa-onnx-ort-zipformer-en-2023-06-26

sherpa-onnx-ort-zipformer-en-2023-06-26

Model details

Conversion

Usage with sherpa-onnx

imedemi/
sherpa-onnx-ort-zipformer-en-2023-06-26