IISc Speech AI Research Internship 2026: On-Device TTS and STT for Indian Languages, 3 to 6 Months
IISc Bengaluru's Computational Intelligence Lab wants research interns to build on-device speech models (TTS and STT) for English and Indic languages. Unpaid, 3 to 6 months on campus; apply with a resume, write-up and demo link.
On this page 6 sections
The Computational Intelligence Lab in the Department of Aerospace Engineering at the Indian Institute of Science (IISc), Bengaluru, is hiring research interns to build speech models for Indian users. The work covers text-to-speech (TTS) and speech-to-text (STT) in English and Indic languages, and the models have to run on the phone itself: offline, on cheap hardware, without a fast internet connection. The internship runs for 3 to 6 months at IISc and is unpaid. You apply through a Google Form with your resume, a short write-up of your speech work and a link to a demo.
The lab wants people who have already built something in this area, even as a side project. If you've fine-tuned Whisper on Hindi audio or squeezed a model onto a Raspberry Pi, this is aimed at you.
| Detail | Information |
|---|---|
| Role | Research Intern, Speech AI |
| Lab | Computational Intelligence Lab, Department of Aerospace Engineering, IISc |
| Location | IISc campus, Bengaluru (on site) |
| Duration | 3 to 6 months |
| Stipend | None, this is an unpaid research internship |
| Work | On-device speech recognition and speech synthesis for English and Indic languages |
| Who the form is for | Mainly 3rd-year and final-year students; there is an "Other" option too |
| How to apply | Google Form with resume, write-up and demo link |
| Last date | Not announced, so apply early |
What the IISc speech AI interns will work on
The lab's goal is speech models that are accurate but small enough to run on the devices people in India actually own. That means offline use, low cost and support for Indian languages, not only English. Interns will take models from raw audio data to something that works on a phone or a small board.
In practice, expect three kinds of work:
- Fine-tuning speech models for recognition (ASR), synthesis (TTS) or speech understanding
- Collecting, cleaning and preparing your own audio datasets in English or Indic languages
- Shrinking and deploying those models on edge devices
Skills the lab is looking for
The post lists four things it wants to see:
- Hands-on fine-tuning of speech models such as Whisper, Wav2Vec2, IndicConformer or similar.
- Work with custom datasets: collecting, cleaning and preparing audio, especially in English or an Indic language.
- Edge deployment on mobile phones, Raspberry Pi or Jetson Nano, using quantization, pruning, ONNX, TFLite or similar tools.
- Ownership: being able to work on your own and take a project from raw data to a working on-device model.
What gives you an edge
- Experience with low-resource languages
- Code-mixed speech such as Hinglish or Tanglish
- Real-time or streaming inference
Is the IISc speech AI internship paid?
No. The application form says plainly that this is an unpaid research internship. You'll also need to be in Bengaluru, because the form asks whether you can commit 3 to 6 months to working at IISc. Plan for rent and living costs before you apply. What you get in return is research experience at IISc on a problem that matters, which helps if you're aiming for a master's or a research role later.
If you need a paid option, our guide to research internships in India lists programmes with stipends, and the IISc internships guide covers the institute's other openings.
What the application form asks for
All questions on the form are compulsory:
- A brief write-up of your work on speech models and edge deployment
- A link to a demo of your speech AI work
- Your current year of study (Final year, 3rd year, or Other)
- Whether you plan to do a master's
- Whether you've deployed speech models on edge devices
- Your contact number
- Whether you can commit to the next 3 to 6 months at IISc
- Your resume (one file, up to 10 MB)
The form needs a Google sign-in because it takes a file upload, and it records the name and email of the Google account you use. Shortlisted candidates may be called for a follow-up discussion.
How to apply for the IISc speech AI internship
- Put your best speech project online with a demo people can try or watch: a Hugging Face Space, a GitHub repo with a short screen recording, or an APK that runs offline.
- Write a short note on what you built, which model you started from, what data you used and how you got it to run on a device (model size, latency, the tools you used).
- Update your resume so the speech and edge work sits at the top.
- Fill in the IISc Speech AI Internship application form and upload your resume.
Tips for a stronger application
The demo link matters more than anything else here. A working model that runs offline on a phone, even a rough one, says more than a list of courses. If you only have a cloud notebook, spend a weekend exporting the model to ONNX or TFLite and running it on a device, then report the numbers.
Indian-language work stands out. A Whisper model fine-tuned on Tamil, Marathi or code-mixed Hinglish recordings that you collected and cleaned yourself shows exactly the skills the lab lists. Say how many hours of audio you used and what error rate you reached.
For more openings, browse all opportunities on Talentd.
Radhika is a Technical Content Writer at Talentd, where she creates clear, practical, and research-driven content focused on jobs, internships, career guidance, and technology. She specializes in simplifying complex technical topics and turning them into easy-to-understand resources for students and fresh graduates. At Talentd, Radhika contributes to articles related to hiring updates, interview preparation, career roadmaps, and industry trends, helping thousands of learners stay informed and make better career decisions. Her work focuses on delivering accurate, actionable, and student-friendly content that supports the growing Talentd community.