Thai Audio Segmentation & Transcription

Lifted (Upwork); United States; Remote; USD 17.000 - 28.000; Teilzeit

Zusammenfassung

Lifted is seeking Thai audio segmentation and transcription experts to support an AI training initiative aimed at improving Thai speech recognition models. In this role you will produce precise, verbatim transcripts in Thai, capturing every utterance, including fillers and repetitions, with speaker labels for 2–8 participants and timestamps within 500 ms.

You will annotate non-speech events such as laughter or background noise, adhere to project instructions, and maintain steady productivity

Qualifikationen

• Native-level fluency in Thai, including familiarity with regional accents and informal speech.
• Strong listening comprehension and attention to detail.
• Ability to produce verbatim transcripts, including stutters, repetitions, and filler words.
• Experience in transcription, annotation, or linguistics is an advantage.
• Reliable internet connection and access to a computer; mobile devices are not supported.

Aufgaben

• Listen to audio recordings and produce verbatim transcripts in Thai.
• Capture how it is said, including filler words and repetitions.
• Add speaker labels for multi-speaker recordings (2–8 participants).
• Insert timestamps for segments (within 500 ms accuracy).
• Annotate non-speech events such as laughter, background noise, or unclear audio.
• Follow detailed project instructions and maintain high quality standards.

Kenntnisse

Thai fluency

Listening skills

Verbatim transcription

Attention to detail

Remote work readiness

Tools

Browser tools

Jobbeschreibung

• Listen to audio recordings and produce accurate, verbatim transcriptions in Thai.


• Capture not only what is said, but how it is said (including filler words, repetitions, and false starts).


• Add speaker labels for multi-speaker recordings (2-8 participants).


• Insert precise timestamps for text segments (within 500ms accuracy).


• Annotate non-speech events such as laughter, background noise, or unclear audio.


• Follow detailed project instructions and maintain high quality standards.


• Maintain steady productivity and reliability throughout the project.

Qualifications

• Native-level fluency in Thai, including familiarity with regional accents and informal speech.


• Strong listening comprehension and attention to detail.


• Ability to produce verbatim transcriptions, including stutters, repetitions, and filler words.


• Experience in transcription, annotation, or linguistics is an advantage.


• Comfortable using browser-based tools (training will be provided).


• Reliable internet connection.


• Access to a laptop or desktop computer (mobile devices are not supported).


• Reliable availability for 20+ hours per week.

Additional information

• Work anytime, provided you meet project deadlines and quality standards.


• Tasks and overall workload may shift based on project needs.


• Consistent availability and productivity are required throughout the assignment.


• Payment will be based on your completed output.

About Lifted

Our enterprise client, a leading provider of high-quality, diverse datasets for AI model development is looking for a Thai Audio Segmentation and Transcription experts to support an AI training initiative aimed at improving Thai speech recognition models.

In this role, you will produce precise, verbatim transcriptions and review audio recordings in Thai. Your contributions will directly help train and enhance next-generation AI speech technologies.

Work Assignment: This project focuses on transcription tasks involving both single-speaker and multi-speaker audio recordings. Tasks may vary in complexity depending on audio quality and number of speakers.

Task Summary (What to Expect)

You will listen to real audio recordings and produce precise, verbatim transcripts - capturing not just what was said, but how it was said. That means every \"uh,\" every repeated word, and every trailing sentence matters. Your work will include:

• Verbatim transcription with a target of 98%+ accuracy


• Timestamping segments with less than 500ms deviation


• Speaker labeling for conversations with 2 to 8 participants


• Event annotation for non-speech elements (e.g., laughter, background noise, unintelligible audio)