Speech Quality Evaluation Tutorial

Annotation Guideline

Purpose

Rate the audio clip using three 1-5 scores:

Rate only what you hear. Do not judge based on any outside information.

Listening Instructions

Use headphones if possible, listen in a quiet place, and keep the volume at a comfortable level.

Rating Scale

ScoreMeaning
5Excellent
4Good
3Fair / acceptable
2Poor
1Bad

1. Naturalness

Question

How natural and human-like does the speech sound?

Focus on:

ScoreGuideline
5Very natural; sounds like a real human speaker.
4Mostly natural, with only small unnatural moments.
3Acceptable, but clearly synthetic or slightly awkward.
2Unnatural, robotic, or distracting.
1Very unnatural, broken, or hard to listen to.

2. Audio Quality

Question

How clean and high-quality is the audio?

Focus on:

Do not judge whether the voice sounds human here. Only judge the technical sound quality.

ScoreGuideline
5Clean, clear audio with no noticeable problems.
4Good audio with minor issues that are not distracting.
3Noticeable noise or artifacts, but still usable.
2Poor audio quality; problems are distracting.
1Very poor or unusable audio.

3. Intelligibility

Question

How easy is it to understand the words?

Focus on:

ScoreGuideline
5All words are clear and easy to understand.
4Mostly clear, with one or two minor issues.
3Some unclear or incorrect words, but the meaning is understandable.
2Hard to understand; several words are unclear or wrong.
1Mostly unintelligible or incorrect.
Start contributing!