Audio AI Intern — Speech Data Labelling Unpaid internship · 2 months · 4 hours a day, 5 days a week · Any degree · Third year preferred · Immediate start Help a speech recognition system learn to understand children — by checking what it hears against what they really said. About the work HOPE is an audio AI project: a speaking-practice companion that talks with school children and listens to their replies. Getting a machine to understand a nine-year-old is genuinely hard — young voices are quiet, sentences slip between Telugu, Hindi and English, and classrooms are loud. So the system's first attempt at writing down what it heard is frequently wrong. Your job is to put that right. Every correction you make becomes training data, and the model learns from it. This is the part of machine learning that rarely appears in a syllabus: no system performs better than the material it was taught on, and building that material is the work. What you would do Listen to short recordings of school children, usually 10 to 30 seconds each. Compare the agent's text against what you hear, and mark each one Correct, Incorrect, No transcript or Unclear. Where it is wrong, type what the child actually said. Work through recordings class by class, and report recurring problems such as audio that is always too quiet. Keep every recording and every child's details confidential. THE MOST IMPORTANT PART OF THE JOB Write the words the child said, never the English meaning. Type them in the local script or in English letters — both are fine. CHILD SPEAKS WRITE EITHER Telugu Hindi నాకు మామిడి ఇషం naaku mamidi istam NEVER I like mangoes मुझे आम पसंद ह mujhe aam pasand hai I like mangoes What you need Fluency in Telugu or Hindi — these are the languages the children speak. No special keyboard or script-typing skill is needed; you may write in the language's own script or in English letters. Fluency in English — required of every labeller: some children speak English, and the tool is in English. Attention to detail and patience. Accuracy matters, speed does not: a wrong transcript is worse than none, and marking a clip Unclear is better than guessing. Good hearing, headphones, a reliable internet connection, and comfort using a web browser. Willingness to handle children's recordings confidentially. This is an unpaid internship lasting two months, with working hours of four hours a day, five days a week. We are looking for ten interns — five fluent in Telugu and five fluent in Hindi. Every intern needs English as well. Students of any degree are eligible. We especially welcome third-year students who can start immediately. Helpful but not required: experience in transcription, translation, teaching, or working with children. What you take away You will finish with a working understanding of how speech datasets are built and judged — the groundwork behind every voice AI system. Training No technical background needed. Before you start, we show you how to use the review page, how to choose between the four options above, and how to write a correction. You also get a short written guide to keep. Everything runs in a web browser, so there is nothing to install

Audio AI Intern — Speech Data Labelling

Emma

Responsibilities

Skills wanted:

Language(s) required:

Akhila Palle

About ApexNeural Private Limited:

mission:

apexneural.com/

Admin access needed