Pulse — Speech-to-Text by Smallest AI

Pulse, built by Smallest AI, is a speech-to-text API purpose-built for real-time voice applications: live call transcription, voice agents, and captioning, alongside a high-throughput batch mode for recorded audio. It streams over WebSocket with sub-100ms time-to-first-token at low concurrency, and includes speaker diarization, word- and sentence-level timestamps, keyword boosting for domain vocabulary, voice-activity events, and inverse text normalization out of the box.

A companion model, Pulse Pro, trades multilingual support for English-only batch transcription tied for #2 on the public Open ASR Leaderboard, built for high-volume call center, meeting, and financial-audio transcription at scale.

Both models include built-in PII and PCI redaction (currently English and Hindi), and the platform holds ISO 27001, SOC 2 Type II, HIPAA, and GDPR compliance, with on-premise deployment available on Enterprise plans.

Key Features

Real-Time Streaming Transcription

• Sub-100ms time-to-first-token at 1 concurrency for live voice AI

• Streams over WebSocket with partial and final transcripts as audio arrives

Multilingual Speech Recognition

• 21 languages on streaming, 26 languages on batch

• Regional auto-detect and mid-session code-switching

Speaker Diarization and Timestamps

• Automatic speaker diarization with per-word and per-utterance labels

• Word- and sentence-level timestamps with per-word confidence scores

Compliance-Ready Redaction

• Built-in PII and PCI redaction, no preprocessing pipeline required

• Supports regulated industries: healthcare, finance, debt collection

Custom Vocabulary and Voice Intelligence

• Keyword boosting: up to 10,000 custom terms per session for brand names, jargon, and product terms

• Emotion and gender detection on batch transcription

Pulse Pro: Leaderboard Accuracy

• English-only, batch-only speech-to-text model

• Tied for #2 on the public Open ASR Leaderboard

Use Cases

• Voice agents and conversational AI: real-time transcription for live voice bots

• Call center and contact center analytics: transcribe and analyze recorded calls at scale

• Meeting and financial-audio transcription: high-accuracy batch processing for enterprise audio

• Multilingual customer support: transcribe support calls across 20+ languages

• Regulated industries: HIPAA and PCI-compliant transcription with built-in redaction

Security and Compliance

Pulse is backed by Smallest AI's enterprise-grade trust center, covering ISO 27001, SOC 2 Type II, HIPAA, and GDPR certifications, with on-premise deployment available for teams that need full data residency.

FAQ

What languages does Pulse support?

Pulse supports 21 languages on real-time streaming and 26 languages on batch transcription, including English, Hindi, Spanish, French, German, Mandarin, Japanese, and Korean, with regional auto-detect for unknown audio.

What is the difference between Pulse and Pulse Pro?

Pulse handles both streaming and batch transcription across 20+ languages. Pulse Pro is a batch-only, English-only model tuned for maximum accuracy, tied for #2 on the public Open ASR Leaderboard. Use Pulse for live voice agents and multilingual audio, and Pulse Pro for high-volume English transcription where accuracy matters most.

Does Pulse support real-time transcription?

Yes. Pulse streams transcription results over WebSocket with sub-100ms time-to-first-token at low concurrency, making it suitable for live voice agents, real-time captioning, and conversational AI.

Is Pulse HIPAA and GDPR compliant?

Yes. The Smallest AI platform holds ISO 27001, SOC 2 Type II, HIPAA, and GDPR certifications, and offers built-in PII and PCI redaction for regulated workflows in healthcare, finance, and debt collection.

Can Pulse redact sensitive information automatically?

Yes. Pulse includes built-in PII and PCI redaction with no separate preprocessing pipeline required. Redaction currently covers English and Hindi audio.

What is keyword boosting used for?

Keyword boosting lets you supply up to 10,000 custom terms per session, such as brand names, product names, or domain jargon, so Pulse recognizes them more accurately in the transcript.

Does Pulse offer on-premise deployment?

Yes. On-premise deployment is available on Smallest AI's Enterprise plan for teams that require full data residency and control.

Does Pulse detect emotion or speaker gender?

Yes, on batch transcription. Pulse can return emotion detection and gender detection as optional outputs alongside the transcript.

Learn more or start building at smallest.ai/speech-to-text.