Omni SenseVoice 🚀

The Ultimate Speech Recognition Solution

Built on SenseVoice, Omni SenseVoice is optimized for lightning-fast inference and precise timestamps—giving you a smarter, faster way to handle audio transcription!

Install

pip3 install OmniSenseVoice

Usage

omnisense transcribe [OPTIONS] AUDIO_PATH

Key Options:

--language: Automatically detect the language or specify (auto, zh, en, yue, ja, ko).
--textnorm: Choose whether to apply inverse text normalization (withitn for inverse normalized or woitn for raw).
--device-id: Run on a specific GPU (default: -1 for CPU).
--quantize: Use a quantized model for faster processing.
--help: Display detailed help information.

Benchmark

omnisense benchmark -s -d --num-workers 2 --device-id 0 --batch-size 10 --textnorm woitn --language en benchmark/data/manifests/libritts/libritts_cuts_dev-clean.jsonl

Optimize	test set	GPU	WER ⬇️	RTF ⬇️	Speed Up 🔥
onnx	dev-clean[:100]	NVIDIA L4 GPU	4.47%	0.1200	1x
torch	dev-clean[:100]	NVIDIA L4 GPU	5.02%	0.0022	50x
onnx `fix cudnn`	dev-clean[all]	NVIDIA L4 GPU	5.60%	0.0027	50x
torch	dev-clean[all]	NVIDIA L4 GPU	6.39%	0.0019	50x

fix cudnn: cudnn_conv_algo_search: DEFAULT
With Omni SenseVoice, experience up to 50x faster processing without sacrificing accuracy.

# LibriTTS
DIR=benchmark/data
lhotse download libritts -p dev-clean benchmark/dataLibriTTS
lhotse prepare libritts -p dev-clean benchmark/data/LibriTTS/LibriTTS benchmark/data/manifests/libritts

lhotse cut simple --force-eager -r benchmark/data/manifests/libritts/libritts_recordings_dev-clean.jsonl.gz \
    -s benchmark/data/manifests/libritts/libritts_supervisions_dev-clean.jsonl.gz \
    benchmark/data/manifests/libritts/libritts_cuts_dev-clean.jsonl

omnisense benchmark -s -d --num-workers 2 --device-id 0 --batch-size 10 -
-textnorm woitn --language en benchmark/data/manifests/libritts/libritts_cuts_dev-clean.jsonl

omnisense benchmark -s --num-workers 4 --device-id 0 --batch-size 16 --textnorm woitn --language en benchmark/data/manifests/libritts/libritts_cuts_dev-clean.jsonl

Contributing 🙌

Step 1: Code Formatting

Set up pre-commit hooks:

pip install pre-commit==3.6.0
pre-commit install

Step 2: Pull Request

Submit your awesome improvements through a PR. 😊

lifeiteng/OmniSenseVoice