Pulse | Smallest AI Docs

Pulse is a high-accuracy, low-latency speech-to-text model built for real-time transcription across 39 languages, with streaming and non-streaming support.

64ms

TTFT at 1 concurrency

300ms

TTFT at 100 concurrency

39 Languages

Streaming + Non-streaming

2 Modes

Streaming + Non-streaming

Model Overview


Developed by	Smallest AI
Model type	Speech-to-Text
Languages	39 supported
License	Proprietary
Model format (non-streaming)	`pulse_offline_<lang>_<version>.smlst`
Model format (streaming)	`pulse_streaming_<lang>_<version>.smlst`
Documentation	docs.smallest.ai/waves
Console	console.smallest.ai
Support	support@smallest.ai

Key Capabilities

Real-Time Optimized

Ultra-low latency architecture delivering 64ms TTFT at 1 concurrency and 300ms at 100 concurrent requests — designed for live transcription and conversational AI.

Multi-Language

39 languages supported across streaming and non-streaming modes, with automatic language detection and code-switching within a single session.

PII / PCI Redaction

Built-in redaction of personal and payment card data, enterprise-ready for both streaming and non-streaming use cases.

Speaker Diarization

Automatic multi-speaker identification available across both modes. Streaming diarization is enterprise-ready; non-streaming is available with a cap of 4 speakers.

Noise Reduction

Background noise handling built into the model — enterprise-ready in streaming mode.

Code-Switching

Supports multi-language audio within a single session. Best used by setting the known primary language (e.g. es for Spanish handles English+Spanish automatically).

Performance & Benchmarks

Word Error Rate (WER) by language evaluated on the FLEURS dataset. Lower is better. NA = not available or not supported by that provider.

Evaluation. FLEURS dataset across 32 languages. Competitor numbers sourced from AssemblyAI published benchmarks and Deepgram internal benchmarks.

Streaming

European Languages

Indic Languages

Language	Smallest Pulse	Deepgram Nova 2	Deepgram Nova 3	AssemblyAI Universal	AssemblyAI Uni-3 Pro
English	4.5%	7.9%	6.7%	4.38%	8.43%
Italian	6.47%	10.7%	6.2%	3.29%	5.60%
Spanish	7.22%	8.6%	4.1%	2.95%	7.63%
Portuguese	12.41%	9.9%	7.5%	4.80%	7.88%
German	13.05%	8.2%	8.5%	4.99%	11.79%
French	14.29%	13.3%	10.7%	7.71%	9.59%
Russian	16.27%	7.9%	11.8%	5.80%	NA
Ukrainian	NA	12.4%	NA	7.42%	NA
Polish	NA	12.2%	NA	6.63%	NA
Dutch	16.54%	16.3%	12.5%	7.79%	NA
Czech	12.4%	22.9%	19.2%	NA	NA
Slovak	13.5%	31.2%	NA	NA	NA
Swedish	18.7%	17.7%	14.3%	NA	NA
Finnish	18.3%	14.1%	13.2%	10.10%	NA
Latvian	16.5%	48.7%	NA	NA	NA
Romanian	17.8%	36.0%	NA	NA	NA
Estonian	17.8%	49.0%	NA	NA	NA
Bulgarian	24.1%	32.7%	NA	NA	NA
Danish	19.8%	21.1%	16.1%	NA	NA
Hungarian	22.5%	31.8%	28.6%	NA	NA
Maltese	25.5%	NA	NA	NA	NA
Lithuanian	25.1%	44.9%	NA	NA	NA

Pre-recorded

European Languages

Indic Languages

Language	Smallest Pulse	Deepgram Nova 2	Deepgram Nova 3	AssemblyAI Universal
English	4.5%	7.9%	6.7%	8.0%
Italian	3.0%	10.7%	6.2%	NA
Spanish	3.2%	8.6%	4.1%	NA
Portuguese	5.0%	9.9%	7.5%	NA
German	6.4%	8.2%	8.5%	NA
French	7.1%	13.3%	10.7%	NA
Russian	9.6%	7.9%	11.8%	NA
Ukrainian	7.5%	12.4%	NA	NA
Polish	10.3%	12.2%	NA	NA
Dutch	15.0%	16.3%	12.5%	NA
Czech	12.4%	22.9%	19.2%	NA
Slovak	13.5%	31.2%	NA	NA
Swedish	18.7%	17.7%	14.3%	NA
Finnish	18.3%	14.1%	13.2%	NA
Latvian	16.5%	48.7%	NA	NA
Romanian	17.8%	36.0%	NA	NA
Estonian	17.8%	49.0%	NA	NA
Bulgarian	24.1%	32.7%	NA	NA
Danish	19.8%	21.1%	16.1%	NA
Hungarian	22.5%	31.8%	28.6%	NA
Maltese	25.5%	NA	NA	NA
Lithuanian	25.1%	44.9%	NA	NA

Features — Non-streaming

Feature	Available	Notes
Speaker diarization	Yes	Capped at 4 speakers; known issues
PII redaction	Yes	Personal info redaction
PCI redaction	Yes	Payment card data redaction
Word-level timestamps	Yes	Per-word timing
Sentence-level timestamps	Yes	Requires `word_timestamps=true` to be enabled
Punctuation	Yes	Auto punctuation
Profanity filter	Yes	Explicit content filtering
Language detection	Yes	Auto language ID — not enterprise-ready
Code-switching	Yes	Multi-language in same audio
Noise reduction	Yes	Background noise handling
Emotion, age and gender detection	Yes	Returns the percentage score of detected emotion, age, and gender

Features — Streaming

Feature	Available	Notes
Speaker diarization	Yes	Multi-speaker identification
Keyword boosting	Yes	Custom vocabulary enhancement
PII redaction	Yes	Personal info redaction
PCI redaction	Yes	Payment card data redaction
Word-level timestamps	Yes	Per-word timing
Sentence-level timestamps	Yes	Per-sentence timing
Punctuation	Yes	Auto punctuation
Profanity filter	No	—
Language detection	Yes	Auto language ID
Code-switching	Yes	Multi-language in same audio
Custom vocabulary	No	—
Noise reduction	Yes	Background noise handling

Supported Languages — Non-streaming

Language	Code	Available
English	`en`	Yes
Italian	`it`	Yes
Spanish	`es`	Yes
Portuguese	`pt`	Yes
Hindi	`hi`	Yes
German	`de`	Yes
French	`fr`	Yes
Ukrainian	`uk`	Yes
Russian	`ru`	Yes
Kannada	`kn`	Yes
Malayalam	`ml`	Yes
Polish	`pl`	Yes
Marathi	`mr`	Yes
Gujarati	`gu`	Yes
Czech	`cs`	Yes
Slovak	`sk`	Yes
Telugu	`te`	Yes
Oriya (Odia)	`or`	Yes
Dutch	`nl`	Yes
Bengali	`bn`	Yes
Latvian	`lv`	Yes
Estonian	`et`	Yes
Romanian	`ro`	Yes
Punjabi	`pa`	Yes
Finnish	`fi`	Yes
Swedish	`sv`	Yes
Bulgarian	`bg`	Yes
Tamil	`ta`	Yes
Hungarian	`hu`	Yes
Danish	`da`	Yes
Lithuanian	`lt`	Yes
Maltese	`mt`	Yes
Japanese	`ja`	Yes
Cantonese	`yue`	Yes
Mandarin	`zh`	Yes
Korean	`ko`	Yes
Tagalog	`tl`	Yes
Indonesian	`id`	Yes
Malay	`ms`	Yes

Supported Languages — Streaming

Language	Code	Available
English	`en`	Yes
Italian	`it`	Yes
Spanish	`es`	Yes
Portuguese	`pt`	Yes
Hindi	`hi`	Yes
German	`de`	Yes
French	`fr`	Yes
Ukrainian	`uk`	Yes
Russian	`ru`	Yes
Kannada	`kn`	Yes
Malayalam	`ml`	Yes
Polish	`pl`	Yes
Marathi	`mr`	Yes
Gujarati	`gu`	Yes
Czech	`cs`	Yes
Slovak	`sk`	Yes
Telugu	`te`	Yes
Oriya (Odia)	`or`	Yes
Dutch	`nl`	Yes
Bengali	`bn`	Yes
Latvian	`lv`	Yes
Estonian	`et`	Yes
Romanian	`ro`	Yes
Punjabi	`pa`	Yes
Finnish	`fi`	Yes
Swedish	`sv`	Yes
Bulgarian	`bg`	Yes
Tamil	`ta`	Yes
Hungarian	`hu`	Yes
Danish	`da`	Yes
Lithuanian	`lt`	Yes
Maltese	`mt`	Yes
Japanese	`ja`	Yes
Cantonese	`yue`	Yes
Mandarin	`zh`	Yes
Korean	`ko`	Yes
Tagalog	`tl`	Yes
Indonesian	`id`	Yes
Malay	`ms`	Yes

Best Practices

Specify the language parameter when known

When the language of the audio is known in advance, always set it explicitly rather than relying on automatic detection. This yields better transcription accuracy because the model can optimize directly for that language without needing to first identify it.

For example, setting the language parameter to es (Spanish) tells the model to expect Spanish audio, which also handles English+Spanish code-switching scenarios. This produces more accurate outputs compared to using the multi parameter.

Parameter	Use case
`en`	English
`es`	Spanish (handles English+Spanish)
`hi`	Hindi (handles English+Hindi)
`multi`	Unknown or mixed-language audio only

When to use multi:

When the language is truly unknown beforehand
When processing audio from varied or unpredictable sources

Use features only when needed

Enable optional features (diarization, PII redaction, timestamps) only when the use case requires them. Unnecessary features add latency.

Use Cases

Direct use

Real-time call transcription
Voice assistant input
Meeting transcription
Accessibility and captioning
Customer support recording analysis

Downstream use

Multi-turn conversational agents
Voice-to-text pipelines
Telephony and IVR systems
Content indexing and search
Compliance and audit logging

Limitations & Safety

Known Limitations

Accuracy varies across languages. The following gaps are known and actively being addressed:

Hindi — still training on proper nouns and order IDs; not enterprise-ready for non-streaming
Low-resource languages — Kannada, Malayalam, Marathi, Gujarati, Telugu, Oriya, Bengali, Punjabi, Tamil, Japanese, Cantonese, Mandarin, Korean, Tagalog, Indonesian, and Malay are available but not yet enterprise-ready
Language detection (multi) — automatic language identification does not perform reliably enough for production workloads; specify the known language parameter instead
Non-streaming speaker diarization — capped at 4 speakers; known accuracy issues; contact support for higher speaker count requirements
Audio quality — transcription accuracy is directly affected by input audio quality; background noise, low bitrate, or overlapping speech may degrade results even with noise reduction enabled
Code-switching — works best when the primary language is explicitly set; fully automatic multi-language detection in a single audio stream is not enterprise-ready

Safety & Compliance

Pulse must not be used for:

Recording or transcribing individuals without their explicit consent
Surveillance, stalking, or any form of unauthorized monitoring
Any illegal or unethical purposes

Additionally:

Usage is monitored for policy compliance
For compliance documentation (GDPR, SOC2, HIPAA), contact support@smallest.ai

Contact


Support	support@smallest.ai
Documentation	docs.smallest.ai/waves
Console	console.smallest.ai