---

Automatic-Speech-Recognition-ASR-ONNX

Harness the power of ONNX Runtime to transcribe audio into text effortlessly.

Supported Models

1. Single Model:

   • SenseVoiceSmall
   • Whisper-Large-V3, Whisper-Large-V3-Turbo, Fine-Tune-Whisper-V3-Japanese...
   • Whisper-Large-V2, Fine-Tune-Whisper-V2-Japanese...

2. Combined Models (ASR + Speaker Identify):

   • SenseVoiceSmall + ERes2NetV2
   • SenseVoiceSmall + ERes2NetV2_w24s4ep4

Features

• End-to-end speech recognition with built-in STFT processing.
  Input: Audio file
  Output: Transcription result
• Seamlessly integrate with these additional tools for improved performance:
  • Voice Activity Detection (VAD)
  • Audio Denoiser
• This Whisper does not support automatic language detection. Please specify a target language.

Learn More

• Visit the project overview for further details.

---

性能 Performance

OS	Device	Backend	Model	Real-Time Factor (Chunk Size: 128000 or 8s)
Ubuntu 24.04	Laptop	CPU i5-7300HQ	SenseVoiceSmall f32	0.037
Ubuntu 24.04	Laptop	CPU i5-7300HQ	SenseVoiceSmall q8f32	0.075
Ubuntu 24.04	Desktop	CPU i3-12300	SenseVoiceSmall f32	0.019
Ubuntu 24.04	Desktop	CPU i3-12300	SenseVoiceSmall q8f32	0.022
Ubuntu 24.04	Desktop	CPU i3-12300	SenseVoiceSmall + ERes2NetV2_w24s4ep4 f32	0.1
Ubuntu 24.04	Desktop	CPU i3-12300	Whisper-Large-v3-en q8f32	0.15
Ubuntu 24.04	Desktop	CPU i3-12300	Whisper-Large-v3-Turbo-en q8f32	0.073

---

Coming Soon 🚀

• Paraformer-small-zh

---

自动语音识别（ASR）ONNX

利用 ONNX Runtime 实现音频到文本的高效转录。

支持模型

1. 单模型：

   • SenseVoiceSmall
   • Whisper-Large-V3
   • Whisper-Large-V3-Turbo
   • 自定义微调 Whisper, 例如: kotoba-Japanese
   • Whisper-Large-V3, Whisper-Large-V3-Turbo, 微调-Whisper-V3-Japanese...
   • Whisper-Large-V2, 微调-Whisper-V2-Japanese...

2. 组合模型 (ASR + 讲话者识别)：

   • SenseVoiceSmall + ERes2NetV2
   • SenseVoiceSmall + ERes2NetV2_w24s4ep4

功能特点

• 端到端语音识别，内置 STFT 处理。
  输入：音频文件
  输出：转录结果
• 推荐搭配以下工具，提升性能：
  • 语音活动检测 (VAD)
  • 音频去噪
• 此 Whisper 不支持自动语言检测。请指定目标语言。

了解更多

• 访问项目概览获取更多信息。

---