Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.
Python asraudiochineseemotion-recognitionfunasrmcp-servermultilingual-asropenai-compatible-apiparaformerpunctuationpytorchreal-time-asrspeaker-diarizationspeech-recognitionspeech-to-textstreaming-asrtranscriptionvllmvoice-activity-detectionwhisper-alternative
readme: merge GPU/CPU benchmark into one compact table
游雁d50c2917fa63
diff --git a/README_new_proposal.md b/README_new_proposal.mdindex af94de897..99d846bf8 100644--- a/README_new_proposal.md+++ b/README_new_proposal.md@@ -69,28 +69,17 @@ That's it. **One model, one call** — VAD segmentation, speech recognition, pun## Benchmark-> 7 models on GPU, 3 on CPU. 184 long-form Chinese audio files (192 min total). [Full report →](https://modelscope.github.io/FunASR/benchmark.html)--### GPU Results (PyTorch)--| Model | Type | Speed | Highlights |-|-------|------|-------|-----------|-| **SenseVoice-Small** | NAR | **170x** realtime | Fastest + most accurate |-| **Paraformer-Large** | NAR | **120x** realtime | Best speed/accuracy tradeoff |-| Whisper-large-v3-turbo (OpenAI) | AR | 46x realtime | Fastest Whisper variant |-| Whisper-large-v3-turbo (FunASR) | AR | 26x realtime | With VAD pipeline |-| faster-whisper-large-v3 | AR | 21.5x realtime | CTranslate2 optimized |-| **Fun-ASR-Nano** | LLM | 17x realtime | Best LLM-based accuracy |-| Whisper-large-v3 (OpenAI) | AR | 13.4x realtime | Baseline |--### CPU Results--| Model | Speed | Note |-|-------|-------|------|-| **SenseVoice-Small** | **17.2x** realtime | Production-ready on CPU |-| **Paraformer-Large** | **15.6x** realtime | Production-ready on CPU |-| **Fun-ASR-Nano** | 3.6x realtime | Still faster than realtime |-| Whisper models | - | Impractical on CPU (>2h for 192min audio) |+> 184 long-form audio files (192 min). [Full report →](https://modelscope.github.io/FunASR/benchmark.html)++| Model | GPU Speed | CPU Speed | vs Whisper-large-v3 |+|-------|-----------|-----------|-------------------|+| **SenseVoice-Small** | **170x** realtime | **17x** realtime | 🚀 **13x faster** |+| **Paraformer-Large** | **120x** realtime | **15x** realtime | 🚀 **9x faster** |+| Whisper-large-v3-turbo | 46x realtime | ❌ | 3.4x faster |+| **Fun-ASR-Nano** | 17x realtime | 3.6x realtime | 1.3x faster |+| Whisper-large-v3 | 13x realtime | ❌ | baseline |++> **Key takeaway:** FunASR models run on CPU faster than Whisper runs on GPU.---