Research Intern - Speech AI

Carrerlift; Bengaluru; On-site; INR 223,000 - 391,000; Part time

Job summary

Carrerlift in Bengaluru hosts a 3-6 month research internship with IISc's Computational Intelligence Lab in the Aerospace Engineering department. You will develop speech AI systems for English and Indic languages, aiming for efficient on-device TTS and STT that run offline on common devices.

You will fine-tune Whisper, Wav2Vec2, or IndicConformer, prepare custom audio data, and push end-to-end deployment to edge devices like smartphones or Raspberry Pi.

Qualifications

• Hands-on experience fine-tuning ASR, TTS or speech-understanding systems.
• Practical experience preparing custom audio datasets.
• Prior model deployment on edge hardware.
• Ability to drive a research project end to end independently.

Responsibilities

• Develop and adapt speech-recognition, speech-synthesis or speech-understanding models.
• Fine-tune architectures such as Whisper, Wav2Vec2, IndicConformer or comparable speech models.
• Collect, clean and prepare custom audio datasets for model development.
• Work with speech data in English, Indic languages or both.
• Optimize models for deployment on devices such as smartphones, Raspberry Pi or Jetson Nano.
• Apply compression and deployment techniques including quantization, pruning, ONNX or TFLite.
• Take ownership of the workflow from raw speech data through to a functioning on-device model.

Skills

Fine-tuning ASR/TTS systems

Dataset preparation

Edge deployment

Independent research

Tools

Whisper

Wav2Vec2

IndicConformer

Job description

About the role:

• Join the Computational Intelligence Lab within IISc's Department of Aerospace Engineering for a 3-6 month research internship.
• The project focuses on speech AI systems designed for English and Indic-language users in India.
• The research goal is to create efficient TTS and STT models that can operate directly on commonly available devices.
• The team aims to support offline and low-cost speech applications without depending on high-end hardware or fast connectivity.

What you'll work on:

• Develop and adapt speech-recognition, speech-synthesis or speech-understanding models.
• Fine-tune architectures such as Whisper, Wav2Vec2, IndicConformer or comparable speech models.
• Collect, clean and prepare custom audio datasets for model development.
• Work with speech data in English, Indic languages or both.
• Optimize models for deployment on devices such as smartphones, Raspberry Pi or Jetson Nano.
• Apply compression and deployment techniques including quantization, pruning, ONNX or TFLite.
• Take ownership of the workflow from raw speech data through to a functioning on-device model.

What we're looking for:

• Hands-on experience fine-tuning ASR, TTS or speech-understanding systems is expected.
• Candidates should have practical experience preparing custom audio datasets.
• Prior model deployment on edge hardware is important for the role.
• Applicants should be capable of working independently and driving a research project end to end.

Good to have:

• Experience with low-resource languages is a plus.
• Work involving code-mixed speech such as Hinglish or Tanglish is beneficial.
• Experience with real-time or streaming inference is also valued.
• Relevant side projects are acceptable evidence of practical experience.

Note:

Internship duration is 3-6 months. Candidates with relevant speech AI side projects are encouraged to apply.