Skip to main content

​
1. What is Smallest STT?

Smallest AI Speech-to-Text (STT) is powered by Pulse, a high-accuracy automatic speech recognition engine built for real-time transcription across 31 documented languages. Pulse is designed to deliver sub-100ms first-token latency, making it well-suited for live conversational voice agents where every millisecond of turnaround time matters.

​
2. Key Features of Smallest STT

Pulse offers a set of capabilities tailored for real-time, production voice AI:
  • Ultra-Low Latency: Sub-100ms first-token latency at single concurrency, staying around ~300ms even at 100 concurrent requests — rated for top-tier speed and P95 latency in third-party benchmarks.
  • Broad Language Coverage: Supports 21 individual languages plus regional auto-detect aggregators in streaming mode, including English, Hindi, major European languages, East Asian languages, and South Indian languages (Tamil, Telugu, Kannada, Malayalam).
  • Speaker Identification: Automatic multi-speaker detection with per-word and per-utterance speaker labels.
  • Timestamps and Punctuation: Word- and sentence-level timing information with automatic punctuation insertion.
  • Noise Handling: Background noise processing built into the model itself, reducing the need for separate noise-suppression steps.

​
3. How Bolna Uses Smallest for STT

Bolna AI integrates Smallest AI’s Pulse model to power low-latency transcription for voice agents. Here’s how Bolna leverages Smallest for STT:
  • Real-Time Streaming Transcription: Bolna connects to Pulse’s streaming endpoint to convert caller speech into text with minimal delay, keeping conversations feeling natural and responsive.
  • Multilingual and Code-Mixed Support: With Pulse’s broad language coverage and code-switching support, Bolna voice agents can transcribe conversations that move between English and regional languages within the same call.
  • Speaker-Aware Transcripts: Bolna uses Pulse’s speaker identification to keep agent and caller speech clearly separated in the transcript, improving downstream context handling.
  • Privacy-Conscious Transcription: Pulse’s built-in data redaction helps Bolna avoid persisting sensitive information like personal details or payment card data in call transcripts.

​
4. List of Smallest models supported on Bolna AI

Note: Set model explicitly to pulse when configuring the transcriber — Bolna defaults to nova-2 at the field level if no model is specified.

​
Conclusion

Smallest AI’s Pulse model gives Bolna AI ultra-low-latency, multilingual speech-to-text for real-time voice agents. By integrating Pulse’s speed, code-switching support, and built-in data redaction, Bolna helps businesses deploy responsive, privacy-conscious voice AI across customer service, healthcare, and enterprise automation. For related integrations: