Speech Research Scientist — Bangalore

Team: Core Speech Research
Location: Bangalore, India
Type: Full-time
Experience: No fixed bar — skill and depth matter more than years

About Smallest.ai

Smallest.ai builds real-time voice intelligence systems operating at enterprise scale.
We work across speech recognition, speech generation, and speech-to-speech systems with a strong focus on low latency, multilingual intelligence, and production reliability.

Our goal is simple: Smaller models. Lower latency. Higher intelligence.

Role Overview

As a Speech Research Scientist, you will work on the core speech stack at Smallest.ai.

You will research, train, evaluate, and productionize models across:

• Speech to Text (ASR)

• Text to Speech (TTS)

• Speech to Speech (S2S)

This is not an offline research role.
You will work at the intersection of research, engineering, and real-world deployment.

Core Research Areas

A. Automatic Speech Recognition (ASR)

• Streaming and non-streaming ASR

• Multilingual and code-mixed speech

• Low-latency decoding and inference

• Long-context speech modeling

• Robustness to accents, noise, and telephony audio
  

B. Text to Speech (TTS)

• Neural TTS and generative speech models

• Controllable speech generation including emotion, style, pitch, rate, and prosody

• Speaker adaptation and voice cloning

• Stability, expressiveness, and naturalness optimization
  

C. Speech to Speech (S2S)

• End-to-end speech-to-speech models

• Streaming voice-to-voice architectures

• Codec-based or token-based speech representations

• Low-latency conversational speech generation
  

D. Multilingual and Speaker Intelligence

• Multilingual speaker understanding

• Cross-lingual speaker embeddings

• Speaker identification and verification

• Accent and dialect robustness

• Low-resource language modeling
  

E. Multi-Speaker Modeling

• Multi-speaker diarization

• Overlapping speech detection and separation

• Speaker-aware ASR pipelines

• Joint diarization and recognition modeling

F. Duplex Conversational Models

• Full-duplex speech models

  • Simultaneous listening and speaking

  • Interruption handling and barge-in detection

• Half-duplex conversational models

  • Turn detection

  • Latency-aware response generation

What You Will Build

• Novel model architectures and training strategies

• Large-scale multilingual datasets and pipelines

• Evaluation frameworks for WER, DER, MOS, latency, and RTF

• Streaming inference systems for real-time speech

• Research prototypes converted into production models

• Your work will directly power live customer-facing systems.

Required Skills

• Strong background in speech processing or deep learning

• Deep expertise in at least one of the following:

  • ASR

  • TTS

  • Speech-to-speech systems

• Strong understanding of modern architectures:

  • Transformers, Conformers, diffusion or flow-based models

• Experience with CTC, Transducer, attention-based decoding

• Strong proficiency in PyTorch

• Experience training models at scale

Strong Plus

• Multilingual speech experience (Indic or European languages)

• Speaker embeddings and diarization systems

• Parameter-efficient fine-tuning methods such as LoRA

• Streaming inference optimization

• Deployment experience using ONNX, TensorRT, or Triton

• Publications, open-source contributions, or serious personal research projects

What We Care About

• Depth over buzzwords

• Clean experiments and reproducibility

• Strong benchmarking discipline

• Latency, memory, and throughput awareness

• Research that translates into shipped systems

We value people who ask:

“How does this behave at scale?”

Not just: “Does this work on the dataset?”

Why Smallest.ai

• Work on real-world speech systems at scale

• Direct ownership from research to production

• Close collaboration with founders and infrastructure teams

• Fast iteration cycles with minimal bureaucracy

• Competitive compensation and meaningful ESOPs

• One of the deepest speech research stacks in India

How to Apply

It would be nice if you can also share:

• Resume

• Research papers, GitHub repositories, or technical writing

• Examples of models you trained or systems you built

• A short note on what aspect of LLM or memory research excites you most

Email: [email protected]