Voice Quality Analysis (US English)
Budget: $10 – $30 USD
We are looking for native US English speakers to assist with a Voice Quality Analysis workflow. This project involves evaluating model-generated or fully generated speech for errors to ensure high-quality audio output.
Error Types:
- Audio Glitch – Noise in the audio not part of speech.
- Pronunciation – Incorrect vowel or consonant phonemes.
- Cutoff – Missing text at the beginning or end of audio.
- Hallucination – Extra, repeated, or altered words in audio.
- Intonation – Tone or pitch causes meaning confusion.
- Pausing – Pauses in awkward locations affecting comprehension.
- Inconsistent Speaker Identity – Speaker voice differs from expected identity.
- Inappropriate Persona – Voice emotion distracts from text content.
- Other – Significant errors that do not fit the above categories.
Error Severities:
- Critical – Major errors that disrupt understanding or are highly distracting.
- Medium – Noticeable errors that may negatively impact listener experience.
- Minor – Minor errors that do not affect overall comprehension.
Workflow Steps:
1. Listen to the generated speech fully before evaluating.
2. Highlight specific portions of the text containing errors.
3. Select the type and severity of each error.
4. Use playback controls to verify unclear sections.
5. Submit evaluations with accurate markings for all error instances.
Best Practices:
- Listen to audio multiple times if necessary.
- Be precise when highlighting text and marking errors.
- Consider context for acceptable pronunciation variations.
- Review marked sections before submission.
Start Date: Immediate
Payrate: $1 USD per task for Reviewers and $1.20–$1.50 USD per task for QC
Project Description: https://docs.google.com/document/d/147Nx0yxYLCfp6R4_ynYym0zgW9yjA7rj/edit?usp=sharing&ouid=100871298295076612532&rtpof=true&sd=true
Error Types:
- Audio Glitch – Noise in the audio not part of speech.
- Pronunciation – Incorrect vowel or consonant phonemes.
- Cutoff – Missing text at the beginning or end of audio.
- Hallucination – Extra, repeated, or altered words in audio.
- Intonation – Tone or pitch causes meaning confusion.
- Pausing – Pauses in awkward locations affecting comprehension.
- Inconsistent Speaker Identity – Speaker voice differs from expected identity.
- Inappropriate Persona – Voice emotion distracts from text content.
- Other – Significant errors that do not fit the above categories.
Error Severities:
- Critical – Major errors that disrupt understanding or are highly distracting.
- Medium – Noticeable errors that may negatively impact listener experience.
- Minor – Minor errors that do not affect overall comprehension.
Workflow Steps:
1. Listen to the generated speech fully before evaluating.
2. Highlight specific portions of the text containing errors.
3. Select the type and severity of each error.
4. Use playback controls to verify unclear sections.
5. Submit evaluations with accurate markings for all error instances.
Best Practices:
- Listen to audio multiple times if necessary.
- Be precise when highlighting text and marking errors.
- Consider context for acceptable pronunciation variations.
- Review marked sections before submission.
Start Date: Immediate
Payrate: $1 USD per task for Reviewers and $1.20–$1.50 USD per task for QC
Project Description: https://docs.google.com/document/d/147Nx0yxYLCfp6R4_ynYym0zgW9yjA7rj/edit?usp=sharing&ouid=100871298295076612532&rtpof=true&sd=true