AI Trends

Trending story

Ant's inclusionAI hosts the Ming-Freeform-Audio-Edit benchmark

1 reports1 story updatesUpdated Nov 17 12:41

Understand this story

AI overview

Ant's inclusionAI has published and is hosting the Ming-Freeform-Audio-Edit benchmark dataset on GitHub, for evaluating downstream audio editing tasks of the Ming-UniAudio model. The dataset covers 7 categories of tasks in Chinese and English, including deletion, insertion, and replacement, and provides two instruction forms: Index-based and Content-based. The audio samples come from seed-tts eval, LibriTTS, and Gigaspeech, and are divided into two versions: basic and open. The evaluation script uses WER, editing operation accuracy, and speaker similarity as core metrics, and the audio and meta files need to be downloaded from HuggingFace or ModelScope. There are currently no further developments.

AI-generated from public reports

Latest development · Report

Ant inclusionAI open-sources Ming-Freeform-Audio-Edit audio editing evaluation benchmark

Developments

1 confirmed developments

  1. The inclusionAI repository hosts the Ming-Freeform-Audio-Edit benchmark dataset

    Source:English text is being prepared.

Reports and discussion timeline

Browse reports and discussion by time. Their evidence levels remain distinct.

Nov 17

  1. Ant inclusionAI: New GitHub repositoryFirst-party

    Ant inclusionAI open-sources Ming-Freeform-Audio-Edit audio editing evaluation benchmark

    Ant inclusionAI has released Ming-Freeform-Audio-Edit on GitHub, a benchmark test set for evaluating downstream editing tasks of the Ming-UniAudio model, covering 7 task types across Chinese and English including deletion, insertion, and replacement. The test set is divided into two instruction forms, Index-based and Content-based; audio samples come from seed-tts eval, LibriTTS, and Gigaspeech, and two versions, basic and open, are provided. The evaluation scripts use WER, editing operation accuracy, and speaker similarity as core metrics, and the audio and meta files need to be downloaded from HuggingFace or ModelScope.

Related stories

  1. Related storiesEnglish translation is being prepared.
  2. Related storiesinclusionAI open-sources ZoomGain and releases 4B/9B model checkpoints
  3. Related storiesinclusionAI open-sources the ZwZ model family and the ZoomBench benchmark
  4. Related storiesThe inclusionAI repository provides ZwZ2 training and evaluation code