Audio Captioning Generalist

About the Role

We're hiring careful listeners to review and annotate audio content. You'll transcribe speech, describe non-speech sounds, track speakers through complex conversations, and audit machine-generated drafts for accuracy. The work rewards patience, precision, and a good ear.

What You'll Do

Listen closely to audio clips and produce clear, faithful descriptions of what you hear — speech, voices, music, and ambient sound. You'll separate foreground from background, catch quiet or overlapping details, and keep your judgment consistent across a clip and across a batch. When reviewing machine drafts, you'll flag anything invented, missed, mistimed, or misattributed based strictly on what's audible.

Requirements

You must be a native English speaker with a strong ear and the vocabulary to describe sound precisely. Specifically, you should be able to:

• Listen critically. Hear quiet, brief, overlapping, and changing details, and separate foreground sounds from background ambience.
• Describe what you hear accurately. Write natural English that captures audible evidence clearly and concisely, without adding assumptions the audio doesn't support.
• Track speakers. Attribute every line to the right speaker through overlaps, interruptions, and group speech, and recognize when a new voice enters.
• Transcribe faithfully. Capture English speech including repetitions, fillers, and names; distinguish English accents; and recognize when a line switches to another language.
• Characterize voices. Describe how each voice sounds and how its delivery changes — pitch, voice quality, loudness, pace, emotion, and intonation — comparing voices to each other and judging delivery from the sound rather than the meaning of the words.
• Handle non-speech vocal sounds. Tell apart, count, and time laughs, coughs, sneezes, sobs, gasps, breaths, and similar.
• Describe non-speech sounds. Identify sounds by texture, attack, distance, and relative level; count repeated events; and distinguish discrete events from continuing ambience.
• Describe music in everyday words. What it sounds like, how loud it is relative to other sounds, and when it starts, stops, or changes. No formal music training required.
• Notice foreground shifts. Hear which sounds stand out or cover others, and catch when that changes — including when music starts, stops, or is replaced.
• Judge what belongs to the scene. Tell whether a voice or sound was in the recorded scene or added over it (narration, playback, music bed).
• Place timings accurately. Mark onsets, offsets, speaker changes, overlaps, and gradual transitions.
• Audit machine drafts. Catch invented, missing, mistimed, or misattributed content, relying on what is audible — not on the video, the topic, or how the draft is worded.
• Stay consistent. Apply the same standards throughout a clip and across a batch.