Blanche Ko

Key Skills

Software

Remotasks

Scale AI

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Image

Text

Top Task Types

Audio Recording

Evaluation Rating

Prompt Response Writing SFT

Question Answering

Translation Localization

Freelancer Overview

I bring a strong background in speech data training and multilingual project coordination within the field of AI development. As a native speaker of Traditional Chinese with strong English proficiency, I have hands-on experience in creating and evaluating audio datasets used to train speech models. At Outlier (Scale AI), I worked on high-impact projects involving single- and multi-speaker audio recordings, speech quality evaluation, and classification tasks. Thanks to consistent performance, I was promoted to Senior Reviewer within a month and later appointed Squad Leader, overseeing a team of Traditional Chinese linguists participating in a large-scale conversational AI initiative. My expertise lies in speech data collection and annotation, detailed quality review, and clear communication of feedback to improve model training outcomes. With experience in both labeling and leading diverse linguistic teams, I bring precision, responsibility, and cultural awareness to every project. I’m passionate about training high-quality voice-based AI systems and supporting their development through accurate and scalable data solutions.

Labeling Experience

Translation Quality Review for AI Safety

Scale AITextTranslation LocalizationRLHF

In this project aimed at improving AI safety, my main task was to review and refine AI-generated translations from English to Traditional Chinese, focusing on prompts and responses involving sensitive or potentially harmful topics. I evaluated whether the output aligned with native fluency, cultural context, and safety guidelines. When translations were inaccurate or unnatural, I edited or retranslated the content to ensure clarity, correctness, and appropriateness for native speakers. This work contributed to creating high-quality training data for fine-tuning language models to handle sensitive interactions more responsibly.

In this project aimed at improving AI safety, my main task was to review and refine AI-generated translations from English to Traditional Chinese, focusing on prompts and responses involving sensitive or potentially harmful topics. I evaluated whether the output aligned with native fluency, cultural context, and safety guidelines. When translations were inaccurate or unnatural, I edited or retranslated the content to ensure clarity, correctness, and appropriateness for native speakers. This work contributed to creating high-quality training data for fine-tuning language models to handle sensitive interactions more responsibly.

2024

Audio Prompt Generation for Conversational AI

Scale AIAudioClassificationEmotion Recognition

In this large-scale audio training project, the goal was to develop a speech model capable of understanding and generating more natural, human-like conversations in Chinese. Participants were tasked with creating authentic voice prompts based on predefined conversation types and categories. Each prompt had to simulate real, spontaneous dialogue to ensure the training data reflected genuine human interaction. My specific responsibilities included generating diverse audio prompts in my native language, ensuring natural intonation, clarity, and contextual appropriateness. As a Senior Reviewer and Squad Leader, I also oversaw the quality of submissions from other contributors, conducted prompt evaluations, and provided feedback to maintain data consistency and linguistic standards. The project involved thousands of contributors across different languages, with strict quality checks in place to monitor pronunciation nature, background noise levels, and adherence to conversation context.

In this large-scale audio training project, the goal was to develop a speech model capable of understanding and generating more natural, human-like conversations in Chinese. Participants were tasked with creating authentic voice prompts based on predefined conversation types and categories. Each prompt had to simulate real, spontaneous dialogue to ensure the training data reflected genuine human interaction. My specific responsibilities included generating diverse audio prompts in my native language, ensuring natural intonation, clarity, and contextual appropriateness. As a Senior Reviewer and Squad Leader, I also oversaw the quality of submissions from other contributors, conducted prompt evaluations, and provided feedback to maintain data consistency and linguistic standards. The project involved thousands of contributors across different languages, with strict quality checks in place to monitor pronunciation nature, background noise levels, and adherence to conversation context.

2024

Education

Work History