Companies in this catalog do not charge applicants. Never send money to apply.
ABOUT PERLE
Perle is an AI infrastructure company building expert-driven training data, evaluation systems, and applied AI products for the world's leading labs and enterprises. Our partners include xAI, Samsung, ELM, and Unisys. Headquartered in San Francisco with experts across more than forty markets, we specialize in the work that requires real human judgment: domain expertise, linguistic nuance, and cultural fidelity that generic data vendors cannot deliver.
Apply here:
THE ROLE
We are hiring expert Hebrew (US, Canada, EU) transcribers to convert audio recordings of native speakers into precise, conventionally formatted written transcripts. The work powers speech recognition, dialogue, and evaluation systems for major AI labs. Audio ranges from clean studio recordings to natural conversational speech in real-world acoustic conditions.
You will be working at the level of expert linguistic judgment. This is not data entry. Decisions about orthography, code-switching, speaker attribution, disfluencies, and non-speech events shape the quality of every model trained on your output.
DIALECT AND SCOPE
Modern Hebrew (Ivrit) as spoken across Israel. Project audio reflects everyday speech with right-to-left script direction and awareness of optional niqqud (vowel pointing) per project conventions.
Transcribe audio recordings into accurate, time-aligned written text following Perle's project-specific style guide Apply correct conventions for speaker identification, overlapping speech, disfluencies, fillers, false starts, and non-speech events such as laughter, background noise, and music Make principled orthographic decisions for dialectal, colloquial, and code-switched speech Flag ambiguous segments, uncertain speakers, and audio quality issues using project conventions Self-review every submission against Perle's quality bar before delivery Engage with QA reviewers on edge cases, calibration sessions, and style guide updates MUST HAVE Native fluency in Modern Hebrew with lifelong residence or deep immersion in Israel High accuracy reading and writing in Hebrew script, with or without niqqud per project specification Comfortable transcribing Hebrew-English, Hebrew-Russian, or Hebrew-Arabic code-switching
Familiarity with both colloquial and formal Hebrew registers
At least two years of professional transcription, captioning, court reporting, broadcast, translation, or linguistic annotation experience
Demonstrated accuracy at the word level under deadline pressure
Comfortable with web-based annotation platforms and variable-speed audio playback
Reliable high-speed internet, quality headphones, and a quiet workspace
Ability to commit to defined volume per week during active project phases
NICE TO HAVE
Background in Hebrew linguistics, translation, journalism, or media
Prior experience annotating Hebrew speech data
Working knowledge of Yiddish, Arabic, or Russian for cross-lingual audio
Familiarity with IPA or other phonetic transcription systems
Experience with verbatim versus clean-verbatim transcription conventions
HOW WE WORK WITH YOU
Paid pilot task before full onboarding so both sides can calibrate
Per-task or per-minute compensation, with rates tied to audio complexity and turnaround
Flexible scheduling, with volume scaling up and down based on active client projects
Direct working relationship with Perle's linguistic leads and QA team
Eligible for invitation to additional Perle expert projects across related domains
HOW TO APPLY
Apply through Winnow, Perle's job portal at winnow.perle.ai. Find the role for your language under Open Jobs, set up your candidate profile, and complete the fully automated interview, which takes about 25 minutes. We review applications within 24 hours and invite passing candidates straight to the project.
Apply here:
EQUAL OPPORTUNITY
Perle hires experts on the basis of skill, language proficiency, and demonstrated quality. We welcome applicants from every background and every region where our target languages are spoken.