遇见数据集

Trelis/eval-parakeet-tdt-0.6b-v3-multimed-hard-20260408-1930

收藏
Hugging Face2026-04-08 更新2026-04-12 收录
官方服务:

资源简介:

--- tags: - whisper - evaluation - speech - speech-to-text --- # Evaluation Results: parakeet-tdt-0.6b-v3 Evaluation results from Whisper model evaluation. ## Summary | Model | WER | CER | |-------|-----|-----| | [nvidia/parakeet-tdt-0.6b-v3](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3) | 15.94% | 10.13% | ## Source Data - **Evaluation Dataset:** [Trelis/multimed-hard](https://huggingface.co/datasets/Trelis/multimed-hard) - **Model Evaluated:** [nvidia/parakeet-tdt-0.6b-v3](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3) ## Columns | Column | Description | |--------|-------------| | `audio` | Audio sample (if available from source dataset) | | `reference` | Ground truth transcription | | `prediction` | Model prediction | | `wer` | Word Error Rate for this sample | | `cer` | Character Error Rate for this sample | | `entities` | Entity annotations from source dataset | | `entity_cer` | Per-sample entity CER (-1.0 if no entities) | ## Entity CER **Overall Entity CER:** 23.30% | Category | CER | |----------|-----| | anatomy | 26.38% | | biomarker | 20.77% | | condition | 18.79% | | drug | 0.00% | | organisation | 26.73% | | procedure | 36.69% | --- *Generated by [Trelis Studio](https://studio.trelis.com)*

标签: - Whisper - 模型评估 - 语音 - 语音转文字 # 评估结果:parakeet-tdt-0.6b-v3 本评估结果由Whisper模型评估得到。 ## 评估总结 | 模型 | 词错误率(Word Error Rate, WER) | 字符错误率(Character Error Rate, CER) | |-------|-----|-----| | [nvidia/parakeet-tdt-0.6b-v3](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3) | 15.94% | 10.13% | ## 源数据 - **评估数据集:** [Trelis/multimed-hard](https://huggingface.co/datasets/Trelis/multimed-hard) - **待评估模型:** [nvidia/parakeet-tdt-0.6b-v3](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3) ## 数据列说明 | 列名 | 描述 | |--------|-------------| | `audio` | 音频样本(若源数据集提供) | | `reference` | 基准转录文本 | | `prediction` | 模型预测转录结果 | | `wer` | 单样本词错误率(Word Error Rate, WER) | | `cer` | 单样本字符错误率(Character Error Rate, CER) | | `entities` | 源数据集自带的实体标注 | | `entity_cer` | 单样本实体字符错误率(无实体时取值为-1.0) | ## 实体字符错误率 **整体实体字符错误率:** 23.30% | 类别 | CER | |----------|-----| | 解剖学 | 26.38% | | 生物标志物 | 20.77% | | 病症 | 18.79% | | 药物 | 0.00% | | 组织机构 | 26.73% | | 医疗操作 | 36.69% | *由 [Trelis Studio](https://studio.trelis.com) 生成*

提供机构:
Trelis
二维码
社区交流群
二维码
科研交流群
商业服务