GTR-Voice Subjective Evaluation

Thank you for participating in this subjective evaluation.

Instructions:

1. You will hear 18~19 groups of voices, each group contains a reference voice and three test voices.

2. The text content of the reference voice is different from the text content of the test voice.

3. Click the triangle button on the left side of the audio block with the mouse to play the audio.

4. You need to choose one of the three test voices that is most similar to the reference voice.

5. Based on the voice you selected that is most similar to the reference, you need to rate its voice quality and naturalness of speech

Most similar might mean:

• The style of the pronunciation/articulation/phonation of the voice is similar

Speech quality rating guide:

• 1: Bad (The speech contains intolerable noise, distortion, or other artifacts, making understanding difficult.)
• 2: Poor (The speech contains significant noise, distortion, or other artifacts, requiring some efforts to understand it.)
• 3: Fair (The speech contains noticeable noise, distortion, or artifacts, but is not hard to understand.)
• 4: Good (The speech contains minor noise, distortion, or artifacts, barely affecting the listening experience.)
• 5: Excellent (The speech does not contain perceptible noise, distortion, or artifacts, and is easy to understand.)

Speech naturalness rating guide:

• 1: Not at all natural (The speech sounds completely artificial)
• 2: Slightly natural (The speech sounds mostly artificial)
• 3: Moderately natural (The speech has both natural and artificial elements)
• 4: Mostly natural (The speech sounds mostly natural)
• 5: Very natural (The speech sounds completely natural, as if spoken by a human)

The test samples may include artistic voice types such as dubbing of movie and drama lines, dubbing of cartoon character roles, etc.

GTR-Voice 主观评估

感谢您参与本次主观评估。

测试说明:

1. 您将听到18~19组语音，每组包含一个参考语音和三个测试语音。

2. 参考语音的文本内容与测试语音的文本内容不同。

3. 使用鼠标点击音频块左侧的三角按钮播放音频。

4. 您需要选择三个测试语音中与参考语音最相似的一个。

5. 基于您选择的与参考语音最相似的语音，您需要对其声音质量和语音自然度进行评分。

最相似可能意味着:

• 语音的吐字发音风格相似。

声音质量评分指南:

• 1: 差 (语音包含难以容忍的噪声、失真或其他干扰，理解困难。)
• 2: 较差 (语音包含明显的噪声、失真或其他干扰，需要一些努力才能理解。)
• 3: 一般 (语音包含可察觉的噪声、失真或干扰，但不难理解。)
• 4: 好 (语音包含轻微的噪声、失真或干扰，比较容易理解。)
• 5: 非常好 (语音不包含可感知的噪声、失真或干扰，易于理解。)

语音自然度评分指南:

• 1: 完全不自然 (语音听起来完全是人工合成的。)
• 2: 稍微自然 (语音大部分听起来是人工合成的。)
• 3: 中等自然 (语音既有自然的元素也有人工合成的元素。)
• 4: 大部分自然 (语音大部分听起来是自然的。)
• 5: 非常自然 (语音听起来完全自然，就如同一个真人在说话。)

测试样本中可能包含艺术化声音类型比如影视戏剧台词配音，卡通人物角色配音等。