Human Transcribers Needed – 7 Rare Languages

# Human Transcribers Needed – 7 Rare Languages *(Suggested Upwork title field — 48 characters. Languages: Pashto, Tatar, Welsh, Fulah, Kurdish, Quechua, Sesotho — listed in the body below.)* **Pay: $5 per audio hour transcribed (paid per accepted file)** **Max 10 audio hours per day, per transcriber** **Project window: Rolling submissions, July 20 – September 14, 2026** We're looking for native speakers to transcribe conversational audio for an AI training dataset. This is detail-oriented, listen-and-type work — not spoken recording — done from your own computer, on your own schedule (within a daily hour cap, see below). ## Languages Needed - Pashto - Tatar - Welsh - Fulah - Kurdish - Quechua - Sesotho ## What You'll Do You'll work in one of two modes, depending on the file: - **Correct mode:** You get an audio file + a rough machine-generated draft transcript. You listen, fix every error, correct speaker labels, and adjust timestamps. - **Transcribe mode:** For some languages (often Pashto, Fulah, Quechua, Tatar), the machine draft isn't usable — you transcribe from scratch. Both modes pay the same rate. You'll be told which mode applies per file. For each file, you will: 1. Listen to the full audio. 2. Transcribing the recorded Audio into words 3. Verify every speaker label and timestamp. 4. Self-check against our quality bar before submitting. 5. Attach an attestation confirming the transcript is human-generated or human-QA'd. ## ⚠️ Quality Bar — Please Read Before Applying This is the part that determines whether your work gets accepted and paid, so read carefully: - **Minimum 95% word-level precision** — we spot-check 10-15% of your submissions. - **Verbatim, not clean.** Keep filler words (uh, um), false starts, and repetitions exactly as spoken. Do not paraphrase or fix grammar. - **No translating.** If the speaker said it in Kurdish, the transcript is in Kurdish — never English. - **Every utterance gets a Speaker ID** from the file metadata (not "Speaker 1" — the actual ID). - **Timestamps accurate to within 0.5 seconds**, decimal precision to 2 places. - **No machine-only transcripts.** Every file must be listened to and corrected/verified by you personally. Machine-only submissions are automatically rejected and are grounds for removal from the project. - **Attestation field is required** on every submission: either `"human-generated"` or `"human-QA-ed"`. **Common rejection reasons:** precision below 95%, missing/incorrect attestation, mismatched speaker labels, uncorrected machine timestamp drift, stripped-out fillers, or malformed JSON. Rejected files can be corrected and resubmitted once — repeated rejections may affect your rate or continued participation. ## Output Format (JSON) Every transcript is delivered as structured JSON with: - `file_id`, `language`, `duration_seconds`, `speakers` - An `utterances` array, each with `start`, `end`, `speaker`, `text` (verbatim), and your own `confidence` score (0.0–1.0) - An `attestation` field (`"human-generated"` or `"human-QA-ed"`) Non-verbal sounds are marked in brackets (e.g. `[laughs]`, `[pause]`), numbers are written out as spoken, and overlapping speech is split into separate utterances per speaker. Full format details and examples provided after hiring. ## Pay & Workload - **$5 per audio hour** transcribed, paid per accepted JSON submission. - **Maximum 10 audio hours per day** per transcriber. - Payment is based on audio duration, not how long the transcription takes you. - Accepted = passes our internal QA spot-check. - We expect steady, consistent weekly throughput rather than large batches submitted all at once. ## Who We're Looking For - Native or fluent-native speaker of Pashto, Tatar, Welsh, Fulah, Kurdish, Quechua, or Sesotho - Strong written literacy in the language (spelling, grammar conventions) - Comfortable working with structured JSON output (we'll walk you through the format) - Detail-oriented — this work is about precision, not speed - Reliable, steady availability over the coming weeks ## How to Apply Please include: 1. Which language(s) you can transcribe natively 2. Any prior transcription, captioning, or translation experience 3. Your general daily/weekly availability We're hiring multiple transcribers per language, so first-time transcribers are welcome to apply.